A Novel Communication System of High Capacity

Abstract

Two inventive contributions are made for improvement of communications systems. A first derives the channel capacity of a Time-Limited (TL) system across a communications channel contaminated by interference and by noise. The potential increase in channel capacity compared to current communications systems is due to the availability of an arbitrarily large number of Degrees of Freedom (DOF) with finite access Time (FAT) in a TL system. A second takes advantage of the theory established in the first objective to design novel systems, referred to as Mask-Matched TL systems with FAT DOF, or MTF systems for short. The disclosure shows several embodiments of MTF systems where it is possible to improve the capacity of current communications systems, without having to modify or alter their Power Spectral Density, merely by taking advantage of their existing but unexploited FAT DOF through the 3 MTF design steps introduced in this disclosure.

Claims

1. A method for transmitting frames of information (101, 201, 301) across a communications channel (106, 306, 406), the method comprising; a first conversion operation (102, 202, 302) for converting the frames of information (101, 201, 301) into a discrete-time Time-Limited signal (103, 203, 303) where the frames of information (101, 201, 301) are contained in a plurality of Finite Access Time (FAT) Degrees Of Freedom (DOF) (1307); a second conversion operation (502) for converting the discrete-time Time-Limited signal (103, 203, 303) into a continuous-time signal; and a transmission operation (104, 204, 304) for transmitting the continuous-time signal (103, 203, 303) across the communications channel (106, 305, 406).

2. The method in claim 1, in which the first conversion operation (102, 202, 302) uses a matrix that is designed such that the plurality of FAT DOF is selected (1303) as per MTF Design Step I (1302).

3. The method in claim 2, in which the matrix is Block Toeplitz (1104).

4. The method in claim 2, in which the matrix is further designed such that the selected plurality of FAT DOF (1303) is enhanced (1305) as per MTF Design Step II (1304).

5. The method in claim 4, in which the matrix is further designed such that the selected and enhanced plurality of FAT DOF (1305) is randomized (1307) as per MTF Design Step III (1306).

6. The method in claim 5, in which each column of the matrix (1104) is the result of adding (1210) a plurality of sub-columns (1208, 1209); with the first sub-column (1208) corresponding to a function with a Degree of Differenciability (DOD) larger than 0.

7. The method in claim 6, in which the second sub-column (1209) corresponds to a function with a DOD equal to 0.

8. The method in claim 7, in which the elements of the second sub-column (1209) are independent pseudo-random variables.

9. The method in claim 8, in which the first sub-column (1208) is the result of performing a linear convolution operation (1207) between a plurality of pulses (1215, 1216), each pulse with a DOD larger than 0.

10. The method in claim 9, in which the first pulse (1215, 1216) is the result of performing a circular convolution operation (1202, 1205) between a first sub-pulse (1201, 1204) with a DOD larger than 0; and a second sub-pulse (1203, 1206) with a DOD equal to 0.

11. An apparatus for transmitting frames of information (101, 201, 301) across a communications channel (106, 306, 406), the apparatus comprising a first converter for converting the frames of information (101, 201, 301) into a discrete-time Time-Limited signal (103, 203, 303) where the frames of information (101, 201, 301) are contained in a plurality of FAT DOF (1303); a second converter (502) for converting the discrete-time Time-Limited signal (103, 203, 303) into a continuous-time signal (105, 205, 305); and a transmitter (104, 204, 304) for transmitting the continuous-time signal (103, 203, 303) across the communications channel (106, 306, 406).

12. The apparatus in claim 11, in which the first converter (102, 202, 302) uses a matrix that is designed such that the plurality of FAT DOF is selected (1303) as per MTF Design Step I (1302).

13. The apparatus in claim 12, in which the matrix is Block Toeplitz (1104).

14. The apparatus in claim 12, in which the matrix is further designed such that the selected plurality of FAT DOF (1303) is enhanced (1305) as per MTF Design Step II (1304).

15. The apparatus in claim 14, in which the matrix is further designed the selected and enhanced plurality of FAT DOF (1305) is randomized (1307) as per MTF Design Step III (1306).

16. The apparatus in claim 15, in which each column of the Matrix (1104) is the result of adding (1210) a plurality of sub-columns (1208, 1209); with the first sub-column (1208) corresponding to a function with a DOD larger than 0.

17. The apparatus in claim 16, in which the second sub-column (1209) corresponds to a function with a DOD equal to 0.

18. The apparatus in claim 17, in which the elements of the second sub-column (1209) are independent pseudo-random variables.

19. The apparatus in claim 18, in which the first sub-column (1208) is the result of performing a linear convolution operation (1207) between a plurality of pulses (1215, 1216), each pulse with a DOD larger than 0.

20. The apparatus in claim 19, in which the first pulse (1215, 1216) is the result of performing a circular convolution operation (1202, 1205) between a first sub-pulse (1201, 1204) with a DOD larger than 0; and a second sub-pulse (1203, 1206) with a DOD equal to 0.

Description

4 DETAILED DESCRIPTION OF THE DRAWINGS

[0021] The present invention, both as to its organization and manner of operation, may best be understood by reference to the following descriptions, and the accompanying drawings of various embodiments wherein like reference numerals are used throughout the several views, and in which:

[0022] FIG. 1 is a schematic view of an embodiment of a SISO communications link between one TU 314 and one RU 315, over a communications channel 106, the link comprising [0023] MTF Modulator 102, part of the digital side 510, which converts the (possibly Forward Error Correction (FEC) coded) information vector, {right arrow over (α)} 101, into an MTF (digital) vector, {right arrow over (β)} 103, [0024] transmitter, Tx 104, part of the digital side 510 and analog side 511, 513, which converts the MTF (digital) vector, {right arrow over (β)} 103, into a transmitted MTF (analog) signal, x(t) 105, [0025] communications channel 106 which transports the transmitted MTF (analog) signal, x(t) 105 from Tx 104 to be received as y(t) 107 at Rx 108, [0026] receiver Rx 108, part of the analog side 616, 618 and digital side 617, which converts the MTF (analog) signal, y(t) 107, into a received MTF (digital) vector, {right arrow over (r)} 109, and [0027] MTF Detector 110 part of the digital side 617, which detects the (possibly FEC coded) information vector, {right arrow over (α)} 101, as 111 from the received MTF (digital) vector, {right arrow over (r)} 109, using an estimate (possibly using a training sequence), which we refer to as Channel State Information at Receiver (CSIR) 115, of the state of the communications channel 106.

[0028] FIG. 2 is a schematic view of an embodiment of a MU-MISO communications link comprising K TUs 316, . . . , 318, and one RU 317, across a Multidimensional communications channel 306, the TU 316 comprising [0029] MTF Modulator.sub.1 302, part of the digital side 510, which converts the (possibly FEC coded) information vector, {right arrow over (α)}.sub.1 301, into an MTF (digital) vector, {right arrow over (β)}.sub.1 303, [0030] and Tx.sub.1 304, part of the digital side 510 and analog side 511, 513, which converts the MTF (digital) vector, {right arrow over (β)}.sub.1 303, into a transmitted MTF (analog) signal, x.sub.1 (t) 305.

[0031] The TU 318 comprises [0032] MTF Modulator.sub.K 202, part of the digital side 510, which converts the (possibly FEC coded) information vector, {right arrow over (α)}.sub.K 201, into an MTF (digital) vector, {right arrow over (β)}.sub.K 203, part of the digital side 510, [0033] and Tx.sub.K 204, part of the digital side 510 and analog side 511, 513, which converts the MTF (digital) vector, {right arrow over (β)}.sub.K 203, into a transmitted MTF (analog) signal, x.sub.K (t) 205.

[0034] The RU 317 comprises [0035] Rx.sub.1 308, part of the analog side 616, 618 and digital side 617, which converts the MTF (analog) signal, y.sub.1(t) 307, into a received MTF (digital) vector, {right arrow over (r)}.sub.1 309, and [0036] MTF Detector.sub.1 310, MTF Detector 110 part of the digital side 617, which detects the (possibly FEC coded) information vectors, {right arrow over (α)}.sub.1 301, . . . , {right arrow over (α)}.sub.K 201 as 311, . . . , 211 from the received MTF (digital) vector, {right arrow over (r)}.sub.1 309, using CSIR 116, which is an estimate (possibly using a training sequence) of the state of the Multidimensional communications channel 306.

[0037] FIG. 3 is a schematic view of an embodiment of a MU-MIMO communications link comprising K TUs 316, . . . , 318, and N.sub.r non-collaborating RUs 317, . . . , 319, across a Multidimensional communications channel 406, the TU 316 comprising [0038] MTF Modulator.sub.1 302, part of the digital side 510, which converts the (possibly FEC coded) information vector, {right arrow over (α)}.sub.1 301, into an MTF (digital) vector, {right arrow over (β)}.sub.1 303, [0039] and Tx.sub.1 304, part of the digital side 510 and analog side 511, 513, which converts the MTF (digital) vector, {right arrow over (β)}.sub.1 303, into a transmitted MTF (analog) signal, x.sub.1(t) 305.

[0040] The TU 318 comprises [0041] MTF Modulator.sub.K 202, part of the digital side 510, which converts the (possibly FEC coded) information vector, {right arrow over (α)}.sub.K 201, into an MTF (digital) vector, {right arrow over (β)}.sub.K 203, [0042] and Tx.sub.K 204, part of the digital side 510 and analog side 511, 513, which converts the MTF (digital) vector, {right arrow over (β)}.sub.K 203, into a transmitted MTF (analog) signal, x.sub.K (t) 205.

[0043] The RU 317 comprises [0044] Rx.sub.1 308, part of the analog side 616, 618 and digital side 617, which converts the MTF (analog) signal, y.sub.1 (t) 307, into a received MTF (digital) vector, {right arrow over (r)}.sub.1 309, and [0045] MTF Detector.sub.1 310, MTF Detector 110 part of the digital side 617, which detects the (possibly FEC coded) K information vectors {right arrow over (α)}.sub.1 301, . . . , {right arrow over (α)}.sub.K 201 as 311, . . . , 211 from the received MTF (digital) vector, {right arrow over (r)}.sub.1 309, using CSIR 116, which is an estimate (possibly using a training sequence) of the state of the Multidimensional communications channel 406.

[0046] The RU 319 comprises [0047] Rx.sub.N.sub.r, 208, part of the analog side 616, 618 and digital side 617, which converts the MTF (analog) signal, y.sub.N.sub.r (t) 207, into a received MTF (digital) vector, {right arrow over (r)}.sub.N.sub.r 209, and [0048] MTF Detector.sub.N.sub.r 210, MTF Detector 110 part of the digital side 617, which detects the (possibly FEC coded) information vectors {right arrow over (α)}.sub.1 301, . . . , {right arrow over (α)}.sub.K 201 as 311, . . . , .sub.K 211, from the received MTF (digital) vector, {right arrow over (r)}.sub.N.sub.r, 209, using CSIR 117, which is an estimate (possibly using a training sequence) of the state of the communications channel 406.

[0049] FIG. 4 is a schematic view of an embodiment of a MU-MIMO communications link comprising K TUs 316, . . . , 318, and one RU 321 with N.sub.r collaborating Rxs 308, . . . , 208, across a Multidimensional communications channel 406, the TU 316 comprising [0050] MTF Modulator.sub.1 302, part of the digital side 510, which converts the (possibly FEC coded) information vector, {right arrow over (α)}.sub.1 301, into an MTF (digital) vector, {right arrow over (β)}.sub.1 303, [0051] and Tx.sub.1 304, part of the digital side 510 and analog side 511, 513, which converts the MTF (digital) vector, {right arrow over (β)}.sub.1 303, into a transmitted MTF (analog) signal, x.sub.1(t) 305.

[0052] The TU 318 comprises [0053] MTF Modulator.sub.K 202, part of the digital side 510, which converts the (possibly FEC coded) information vector, {right arrow over (α)}.sub.K 201, into an MTF (digital) vector, {right arrow over (β)}.sub.K 203, [0054] and Tx.sub.K 204, part of the digital side 510 and analog side 511, 513, which converts the MTF (digital) vector, {right arrow over (β)}.sub.K 203, into a transmitted MTF (analog) signal, x.sub.K (t) 205.

[0055] The RU 321 comprises [0056] N.sub.r Rx 308, . . . , 208, part of the analog side 616, 618 and digital side 617, which converts the MTF (analog) signals, y.sub.1(t) , . . . , y.sub.N.sub.r(t) 307, . . . , 207 into N.sub.r received MTF (digital) vectors, {right arrow over (r)}.sub.1, . . . {right arrow over (r)}.sub.N.sub.r, 309, . . . , 209 respectively, and [0057] MTF Detector.sub.1 310, MTF Detector 110 part of the digital side 617, which detects the (possibly FEC coded) K information vectors, {right arrow over (α)}.sub.1 301, . . . , {right arrow over (α)}.sub.K 201, as 311, . . . , 211, from the received MTF (digital) vectors {right arrow over (r)}.sub.1 109, . . . , {right arrow over (r)}.sub.N.sub.r 209, using CSIR 118, which is an estimate (possibly using a training sequence) of the state of the channel 406.

[0058] FIG. 5a is a schematic view of an embodiment of Tx 104 (204, 304) for converting a baseband-sampled (consisting of complex samples) or an IF-sampled (consisting of real samples) MTF (digital 510) discrete-time vector, {right arrow over (β)} 103 (203, 303), into an MTF (analog 511) continuous-time RF signal, x(t) 105 (205, 305), to be transmitted, Tx 104 (204, 304) comprising [0059] a Pre-Channel Filter 500 to pre-filter the baseband-sampled or the IF-sampled MTF (digital 510) vector, {right arrow over (β)} 103 (203, 303), based on the knowledge of the statistics of the channel 106 (306, 406) represented by Channel State Information at Transmitter (CSIT) 215, at Tx 104 (204, 304), [0060] a Digital-to-Analog (D-to-A) Converter 502 to convert the pre-channel baseband-sampled or IF-sampled filtered MTF (digital 510) vector 501 into an analog 511 baseband or IF MTF signal 503, [0061] an Up-Converter 504 to convert the analog baseband or IF MTF signal 503 into an up-converted analog 511 RF MTF (passband) signal 505, [0062] a Power Amplifier (PA) 506 to amplify the up-converted analog 511 RF signal 505 into an amplified up-converted analog RF MTF signal 507, and [0063] an Analog Filter 508 to filter the amplified up-converted analog 511 RF MTF signal 507 into a filtered amplified up-converted analog 511 RF MTF signal x(t) 105 (205, 305), which complies with a pre-specified spectral mask 701.

[0064] FIG. 5b is a schematic view of an embodiment of Rx 108 (208, 308) for converting a received MTF (analog 616) continuous-time RF signal, y(t) 107 (207, 307), into a received baseband-sampled (consisting of complex samples) or IF-sampled (consisting of real samples), MTF (digital) discrete-time vector, {right arrow over (r)} 109 (209, 309), to be used for detecting the original (possibly FEC coded) information vector, {right arrow over (α)} 101 (201, 301), Rx 108 (208, 308) comprising [0065] an Analog Filter 607 to filter the received MTF (analog 616) RF signal, y(t) 107 (207, 307), into a filtered received analog RF MTF signal 608 (to filter and excise out-of-band noise and narrowband interferers), [0066] a Low Noise Amplifier (LNA) 609 to amplify the filtered received analog 616 RF MTF signal 608, into an amplified filtered received analog 616 RF MTF signal 610, [0067] a Down-Converter 611 to convert the amplified filtered received analog 616 RF MTF signal 610, into an amplified filtered received analog baseband or IF MTF signal 612, and [0068] an Analog-to-Digital (A-to-D) Converter 613 to convert the amplified filtered received analog 616 baseband or IF MTF signal 612, into a baseband-sampled or IF-sampled digital 617 amplified filtered received MTF vector 614, [0069] a Post-Channel Filter 615 to convert the baseband-sampled or IF-sampled digital 617 amplified filtered received MTF vector 614 into a post-channel filtered baseband-sampled or IF-sampled MTF vector {right arrow over (r)} 109 (209, 309) based on CSIR 115.

[0070] FIG. 6a is a schematic view of an alternative embodiment of Tx 104 for converting an RF-sampled, MTF (digital 510) discrete-time vector, {right arrow over (β)} 103 (203, 303), (consisting of real samples) into an (analog 513) continuous-time RF MTF signal, x(t) 105 (205, 305), to be transmitted, Tx 104 (204, 304) comprising [0071] a Pre-Channel Filter 500 to pre-filter the RF sampled, pre-channel filtered, MTF (digital) vector, {right arrow over (β)} 103 (203, 303), based on CSIT 215 at Tx 104 (204, 304), [0072] a D-to-A Converter 502 to convert the RF sampled, MTF (digital 510) vector 501, into an analog 513 RF MTF signal 503, [0073] a PA 506 to amplify the analog 513 RF MTF signal 503, into an amplified analog 513 RF MTF signal 509, and [0074] an Analog Filter 508 to filter the amplified analog 513 RF MTF signal 507, into a filtered amplified analog 513 RF MTF signal x(t) 105 (205, 305), which complies with a pre-specified spectral mask 701.

[0075] FIG. 6b is a schematic view of an alternative embodiment of Rx 108 (208, 308) for converting a received (analog 618) continuous-time RF MTF signal, y(t) 107 (207, 307), into a received RF MTF sampled, (digital 617) discrete-time MTF vector, {right arrow over (r)} 109 (209, 309), (consisting of real samples) to be used for detecting the original information vector, {right arrow over (α)} 101 (201, 301), Rx 108 208, 308) comprising [0076] an Analog Filter 607 to filter the received (analog 618) RF MTF signal, y(t) 107 (207, 307), into a filtered received analog 618 RF MTF signal 608 (to filter out-of-band noise and narrowband interferers), [0077] an LNA 609 to amplify the filtered received analog 618 RF MTF signal 608, into an amplified filtered received analog 618 RF MTF signal 610, and [0078] an A-to-D Converter 613 to convert the amplified filtered received analog 618 RF MTF signal 612, into an RF sampled digital 617 amplified filtered received MTF vector 614, [0079] a Post-Channel Filter 215 to convert the RF sampled digital 617 amplified filtered received MTF vector 614 into a post-channel filtered RF sampled MTF vector {right arrow over (r)} 109 (209, 309) based on CSIR 115.

[0080] FIG. 7 shows a spectral mask, .sub.WiFi(f), 701 in dBr versus frequency, f-f.sub.c, as an example (among many) of a mask, where f.sub.c. is the carrier frequency. The selected mask, .sub.WiFi(f), is designated for the IEEE802.11 (also known as WiFi) WLAN mask for a 20 MHz band. The mask, .sub.WiFi(f), 701 contains three distinct (non-overlapping) spectral parts: 1. the Occupied Band 705, 2. the Out-of-Band Emissions (DOBE) Band 706, and 3. the Far Out Spurious Emmisions (FOSE) Band 707. Many other mask constraints are available depending on the standard, frequency band and jurisdiction. While different, all masks must generally contain three distinct spectral parts: Occupied Band 705, DOBE Band 706 and FOSE Band 707, similar to the ones described above for .sub.WiFi(f) 701.

[0081] FIG. 8 shows the PSD, .sub.x(t)(f) 802, in dBr versus frequency, f-f.sub.c, of the MTF signal, x(t) 105 (205, 305), where f.sub.c. is the carrier frequency. The MTF signal, x(t) 105 (205, 305), is designed in an attempt to match its PSD, .sub.x(t)(f) 802, with the spectral mask, .sub.WiFi(f) 701, in FIG. 7 as much as possible, however, without exceeding it. The PSD .sub.x(t)(f) 802, contains two distinct spectral parts: [0082] (f) 808, which is the PSD of (f), to be matched with both the Occupied Band 705 and the DOBE Band 706 of the spectral mask, .sub.WiFi(f) 701, as much as possible, but without exceeding it. [0083] (f) 809, which is the PSD of , to be matched with the FOSE Band 707 of the spectral mask, .sub.WiFi(f) 701, as much as possible, but without exceeding it.

[0084] FIG. 9 displays (f) 808 for MTF.sub.3 with a null at the restricted bands as well as across |f|≤2 GHz, when is selected as with a PSD 908 while is selected as defined in (17) with a PSD (f) 809 pre-distorted according to (20). From FIG. 9, one can deduce that it is possible to increase .sub.MTF arbitrarily by increasing f.sub.s.sub.3 regardless of the value of f.sub.c where f.sub.s.sub.3 is the sampling frequency of MTF.sub.3.

[0085] FIG. 10a compares .sub.TL/ in (5) (shown with “.” markers) with .sub.BL/ in (6) (shown with “*” markers) when =.sub.BL=1, =K=1 and the k.sup.th column, {h.sub.ch}.sub.k of h.sub.Ch corresponds to a rectangular pulse ∀k. In FIG. 10a, .sub.TL/ in (5) is illustrated as a number of curves, each curve corresponding to a value of N.sub.min. The selected values are N.sub.min=1, 2, 4, . . . , 1024, with N.sub.min=1 coinciding with .sub.BL/ in (6). .sub.BL/ in (6) contains a “Low” SNR region 1004 and a “High” SNR region 1005. Similarly, .sub.TL/ in (5) contains a “Low” SNR region 1001 and a “High” SNR region 1003. Unlike .sub.BL/ in (6), .sub.TL/ in (5) also contains a medium SNR region 1002, denoted as “Med. SNR” in FIG. 10a where doubling .sub.TL/ in (5) requires a fixed multiple increase in

[00002]$\frac{{\Lambda }_k\mathrm{��}}{N_{\min }}\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{\mathrm{��}}_m}$

since its contribution towards .sub.TL is mostly linear.

[0086] FIG. 10b shows that based on (10), .sub.MTF consists of two medium SNR regions 1006, 1007. As N increases, doubling .sub.MTF initially requires increasing the SNR by a fixed multiple of , which is the first medium SNR region 1006, referred to as “1.sup.st Med. SNR” in FIG. 10b. Then, later on it requires increasing the SNR by a fixed multiple of 2, which is the second medium SNR region 1007, referred to as “2.sup.nd Med. SNR”. FIG. 10b compares .sub.MTF/ based on (10) with .sub.BL/ in (6), where =.sub.BL=1, =1, d=K=1 and in (8) is a rectangular pulse with in (8) 30dBr below . In FIG. 10b, .sub.MTF/ without Constraint 3 is illustrated as a number of curves (with “.” markers), each curve corresponding to a value of N.sub.min.Math..sub.MTF/ with Constraint 3 is illustrated as a number of points (with “square” markers), each point corresponding to a value of N.sub.min. FIG. 10b shows that indeed Constraint 3 maximizes .sub.MTF/ in both medium SNR regions 1006, 1007.

[0087] FIG. 11a shows an embodiment of h 1101 as a block toeplitz matrix. The sup-matrix

[00003]$\left[\begin{array}{c}{h}_{Basic}\\ 0_{\mathrm{��}\left(L-1\right)\times K}\end{array}\right]\in {ℂ}^{M\times K}$

is used as a building block of h 1101 where h.sub.Basic∈.sup.N×K in FIG. 11a comprises sub-blocks with the first

[00004]$.Math.\frac{N}{\mathrm{��}}.Math.$

sub-blocks consisting of rows each, while the last sub-block consisting of ′ rows, i.e. h.sub.Basic=[h.sub.Basic,1 . . . ].sup.T with h.sub.Basic,1∈.sup.d×K 1102 and ∈.sup.d′×K 1103, where

[00005]${\mathrm{��}}^{\prime }\stackrel{\Delta }{=}N\mathrm{modulo}\left(\mathrm{��}\right),N_{\mathrm{��}}\stackrel{\Delta }{=}.Math.\frac{N}{\mathrm{��}}.Math.$

is the ceiling function,

[00006]$.Math.\frac{N}{\mathrm{��}}.Math.$

is the floor function and [.].sup.T denotes a transpose operation.

[0088] FIG. 11b shows an embodiment of h.sub.MTF 1104 as a block toeplitz matrix, where the K TUs 316, 318 are replaced by one equivalent TU 314 with one equivalent h.sub.MTF 1104. The sub-matrix

[00007]${\stackrel{.fwdarw.}{h}}_{MTF}\stackrel{\Delta }{=}\left[\begin{array}{c}{h}_{B\_\mathrm{MTF}}\\ 0_{\mathrm{��}\left(L-1\right)\times K}\end{array}\right]\in {ℂ}^{M\times K}$

is used as a building block of h.sub.MTF 1104 where h.sub.B_MTF in FIG. 11b comprises N.sub.d sub-blocks with the first

[00008]$.Math.\frac{N}{\mathrm{��}}.Math.$

sub-blocks consisting of rows each, while the last sub-block consisting of ′rows, i.e.

[00009]$h_{B\_\mathrm{MTF}}={\left[h_{B\_\mathrm{MTF},1}\mathrm{.Math.}{h}_{B\_\mathrm{MTF},N_d}\right]}^T$

with h.sub.B-MTF,1∈1105 and

[00010]$h_{B\_\mathrm{MTF},N_d}\in {ℂ}^{{\mathrm{��}}^{\prime }\times K}$

1106.

[0089] FIG. 12a displays an embodiment for implementing equations (8) and (9), Equation (9) [0090] convolves {right arrow over (h)}.sub.1,k 1201 with {right arrow over (g)}.sub.1,k 1203 using a circular convolution, denoted by “” 1202, and [0091] convolves 1204 with 1206 using a circular convolution, denoted by “” 1205, [0092] then, it performs −1 linear convolution operations, (h.sub.1,k{right arrow over (g)}.sub.1,k) 1215 * . . . * () 1216, using a linear convolution operator, denoted by “*” 1207, in order to produce 1208.

[0093] Equation (8) adds 1208 that is obtained from equation (9) with 1209 using an adder denoted by “+” 1210 in order to produce {h.sub.B_MTF}.sub.k 1308.

[0094] FIG. 12b displays a preferred embodiment for implementing equations (8) and (9) where all {right arrow over (h)}.sub.l,k ∀l 1201, 1204 in FIG. 12a are selected as {right arrow over (h)}.sub.REC,l,k 1212, 1213. In this case, 1208 in (8) is denoted as 1214.

[0095] FIG. 13 displays a preferred embodiment 1309 for implementing MTF Design Steps I-III, which designs {h.sub.B_MTF}.sub.k 1308 based on Theorem III with the goal of achieving a desired channel capacity, , for a communications channel with (NT.sub.s).sub.min and T.sub.s,max 1301 defined [0096] MTF Design Step I 1302: This step is referred to as the selection step 1302. It accepts N.sub.min and T.sub.s,max 1301, and generates the N selected FAT DOF 1303. [0097] MTF Design Step II 1304: This step is referred to as the enhancement step 1304. It accepts the N selected FAT DOF 1303, and generates the N selected and enhanced FAT DOF 1305. [0098] MTF Design Step III 1306: This step is referred to as the randomization step 1306. It accepts the N selected and enhanced FAT DOF 1305, and generates the N selected, enhanced and randomized FAT DOF 1307.

[0099] The N selected, enhanced and randomized FAT DOF 1307 are used to form the N elements of the k.sup.th column, {h.sub.B_MTF}.sub.k 1308, of h.sub.B_MTF using an inverse transform 1310.

5. DETAILED DESCRIPTION OF INVENTION

5.1 TL Systems

[0100] An information vector, {right arrow over (α)}∈.sup.Q×1 101, 201, 301, consisting of Q (possibly FEC coded) information symbols, can be transmitted by one or several active transmitters, Tx 104, 204, 304, across a communications channel 106, 306, 406 by converting {right arrow over (α)} 101, 201, 301 into a vector, {right arrow over (β)} 103, 203, 303, defined as

[00011]$\begin{array}{cc}\stackrel{.fwdarw.}{\beta }\stackrel{\Delta }{=}h\stackrel{.fwdarw.}{\alpha }& \left(1\right)\end{array}$

using a matrix, h∈.sup.M×Q 1101, where {right arrow over (β)}∈.sup.M×1 103, 203, 303 consists of samples, each of duration T.sub.s for a total duration for {right arrow over (β)} of MT.sub.s. In this disclosure, we select h 1101, to be block toeplitz, i.e. h 1101 is defined as

[00012]$\begin{array}{cc}h\stackrel{\Delta }{=}{\mathrm{toep}}_{\mathrm{��}}\left\{\stackrel{.fwdarw.}{h}\right\}& \left(2\right)\end{array}$

where toep.sub.d{{right arrow over (h)}} is an operator, which forms h 1101 by repeatedly replicating the sub-matrix {right arrow over (h)} to the right L−1 times, while cyclically shifting {right arrow over (h)} down by d rows for every single replica to the right, with

[00013]$L\stackrel{\Delta }{=}.Math.Q/K.Math.,$

defined as the ceiling of Q/K. {right arrow over (h)}∈.sup.M×K is defined as

[00014]$\stackrel{.fwdarw.}{h}\stackrel{\Delta }{=}\left[\begin{array}{c}{h}_{\mathrm{Basic}}\\ 0_{d\left(L-1\right)\times K}\end{array}\right]\mathrm{where}{h}_{\mathrm{Basic}}\in {ℂ}^{N\times K}$

is referred to as the basic building block, while is the all zero (L−1)×K matrix, with ≤N and

[00015]$M\stackrel{\Delta }{=}N+d\left(L-1\right).$

Interpretation of h 1101: Since each column of h 1101 is responsible for transporting one information symbol in {right arrow over (α)} 101, 201, 301, therefore, {right arrow over (β)} 103, 203, 303 in (1) can model the output of a K−user TL system with spreading gain N<∞ with a number, , of desired transmitters (Txs) intended for a receiver, Rx 108, 308, and a number, K.sub.i, of interfering Txs 204, s.t. K=. The k.sup.th active Tx 104, 204, 304 transmits a vector {right arrow over (β)}.sub.k, which transports the set of L symbols, {α.sub.k,α.sub.k+K, . . . , α.sub.k+(L−1)k}, after converting {right arrow over (α)}.sub.k into a continuous-time signal, x.sub.k(t), of finite duration MT.sub.s, with T.sub.s the duration of one sample in {right arrow over (β)} 103, 203, 303.

[0101] Theorem I assumes that: [0102] 1. The k.sup.th Tx 104, 204, 304 transmits x.sub.k(t) 105, 205, 305 subject to Constraint 1:

[00016]$\begin{array}{cc}\mathrm{Constraint}1\text{:}{\int }_{-\infty }^{\infty }{\mathrm{��}}_{x_k\left(t\right)}\left(f\right)\mathrm{df}\le \mathrm{��}\forall k& \end{array}$

where p is the average allocatable transmit power at any Tx 104, 204, 304 and .sub.x.sub.k.sub.(t)(f) 802 is the Power Spectral Density (PSD) of x.sub.k (t) 105, 205, 305. [0103] 2. The k.sup.th Tx 104, 204, 304 transmits x.sub.k(t) 105, 205, 305 using a single antenna across a communication channel where it is received at Rx 108, 208, 308 using a single antenna. The received signal, y(t), is then sampled at Rx 108, 208, 308 at a sampling frequency

[00017]$f_s\stackrel{\Delta }{=}1/T_s,$

to form a discrete-time signal {right arrow over (r)}∈defined as

[00018]$\begin{array}{cc}\stackrel{.fwdarw.}{r}\stackrel{\Delta }{=}{h}_{Ch}\stackrel{.fwdarw.}{\alpha }+\stackrel{.fwdarw.}{w}& \left(3\right)\end{array}$

where h.sub.Ch∈ corresponds to h 1101 after including the effects of the channel 106, 306, 406 such as replacing M by a number ≥M, and {right arrow over (w)}∈ models the WGN. We refer to the combination of the TL system and channel as a TL channel.

[0104] Theorem I: The capacity, .sub.TL, of the TL channel corresponding to h.sub.Ch in (3), subject to Constraint 1, is

[00019]$\begin{array}{cc}{\mathrm{��}}_{TL}=\frac{1}{M{T}_s}{.Math.}_{k=1}^{\mathrm{rank}\left(h_{Ch}\right)}{\log }_2\left(1+{\Lambda }_k\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{f}_s}\right)\mathrm{bps}& \left(4\right)\end{array}$

[0105] where .sub.o/2 is the two-sided PSD of the WGN, is the average attenuation in power across the channel, and Λ.sub.k is the k.sup.th squared singular value of a normalized h.sub.Ch s.t. its k.sup.th column, {h.sub.Ch}.sub.k, has on average an L2-norm, which equals ∀k.

[0106] Importance of Theorem I: .sub.TL in (4) consists of several regions, which depend on the average received TL SNR,

[00020]$\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{f}_s}.$

Similar to the capacity, .sub.BL, of a BL system, which consists of a low SNR region 1004 and a high SNR region 1005, .sub.TL in (4) also consists of a low SNR region 1001 and a high SNR region 1003. Unlike BL systems, .sub.TL in (4) also contains a new medium SNR region 1002, when a number, , of the terms,

[00021]${\Lambda }_k\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{f}_s},$

in (4) are

[00022]$<<1s.t.{\log }_2\left(1+{\Lambda }_k\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{f}_s}\right)\approx {\log }_{2}e{\Lambda }_k\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{f}_s}.$

When .sub.TL is in the low SNR region 1001, =rank (h.sub.Ch). When .sub.TL is in the high SNR region 1003, =0. When .sub.TL is in the medium SNR region 1002, 0<≤rank (h.sub.Ch).

[0107] Given that some of the communication channels in this disclosure are to be constrained by a spectral mask, Theorem I must be modified to include a mask constraint. First, we define the bandwidth (BW) of x.sub.k(t), then, introduce the mask constraint.

[0108] Definition of the BW of x.sub.k(t) 105, 205, 305: Since x.sub.k(t) 105, 205, 305 is TL, its PSD, .sub.x.sub.k.sub.(t)(f) 802, exists over the entire frequency domain f∈{−∞, ∞}, allowing for numerous definitions of BW to exist. In this disclosure, we adopt the same definition of BW as the one adopted by the International Telecommunication Union (ITU), which defines transmitter spectrum emissions as falling into three distinct bands: (a) Occupied band 705 emission with a BW : (b) Out-Of-Band-Emission (DOBE) band 706 with a BW .sub.OOBE; and (c) Far-Out-Spurious-Emission (FOSE) band 707 with an allowable power level

[00023]$\le \frac{1}{ℛ}.$

By adopting the same definition for BW as the ITU, we select the BW, .sub.TL, of the TL system to be defined as the BW, , of the occupied band 705.

[0109] Spectral Mask Constraint 701: Some systems considered in this disclosure are constrained by a spectral mask, .sub.Mask(f) 701. In this case, x.sub.k(t) 105, 205, 305 is subject to Constraint 2:

[00024]$\begin{array}{cc}\mathrm{Constraint}2\text{:}{\mathrm{��}}_{x_k\left(t\right)}\left(f\right)\le \daleth {\mathbb{P}}_{\mathrm{Mask}}\left(f\right)\forall f,\forall k& \end{array}$

where is a normalization constant, which depends on , .sub.mask (f) 701 and h 1101. According to the ITU, must be selected≤.sub.m where .sub.m is the BW of .sub.mask(f) 701. This implies that .sub.TL must be selected≤.sub.m as well. For this reason, we define in this disclosure an overhead factor,

[00025]$\mathrm{��}\stackrel{\Delta }{=}{\mathrm{��}}_{TL}N{T}_s\ge 1,$

as the overhead, both in time and in frequency, which is required for x.sub.k(t) 105, 205, 305 to comply with Constraint 2. It is selected such that .sub.TL≤.sub.m or equivalently, N is selected

[00026]$s.t.N\ge N_{\min }\stackrel{\Delta }{=}\frac{\mathrm{��}}{{\mathrm{��}}_m{T}_s}.$

[0110] Under Constraints 1-2, .sub.TL in (4) can be expressed as

[00027]$\begin{array}{cc}{\mathrm{��}}_{\mathrm{TL}}=\frac{{\mathrm{��}}_m}{\mathrm{��}}\frac{N_{\min }}{M}{.Math.}_{k=1}^{\mathrm{rank}\left(h_{\mathrm{Ch}}\right)}{\log }_2\left(1+\frac{{\Lambda }_k\mathrm{��}}{N_{\min }}\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{\mathrm{��}}_m}\right)\mathrm{bps}& \left(5\right)\end{array}$

[0111] Similarly, under Constraints 1-2, a BL system, of fixed BW, .sub.BL, selected as .sub.BL=.sub.m, has a BL capacity, .sub.BL, given as

[00028]$\begin{array}{cc}{\mathrm{��}}_{BL}=\frac{{\mathrm{��}}_m}{{BL}_{}}{\log }_2\left(1+\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{\mathrm{��}}_m}\right)\mathrm{bps}& \left(6\right)\end{array}$

where .sub.BL is defined as the overhead factor, both in time and in frequency, which is required for the BL system to comply with Constraint 2. When

[00029]$\frac{\stackrel{\_}{\stackrel{\_}{\mathrm{��}}}K\mathrm{��}}{N_o{\mathrm{��}}_m}>1,$

(6) implies that doubling .sub.BL, with a fixed BW, requires a geometric multiple increase in

[00030]$\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{\mathrm{��}}_m}$

since its contribution towards .sub.BL is logarithmic.

[0112] FIG. 10a compares .sub.TL/.sub.m in (5) (shown with “.” markers) with C.sub.BL/.sub.m, in (6) (shown with “*” markers) versus the average received SNR that is normalized w.r.t. .sub.m, when =.sub.BL=1, =K=1 and the k.sup.th column, {h.sub.Ch}.sub.k of h.sub.Ch corresponds to a rectangular pulse ∀k. In FIG. 10a, .sub.TL/.sub.m in (5) is illustrated as a number of curves, each curve corresponding to a value of N.sub.min. The selected values are N.sub.min=1, 2, 4, . . . , 1024, with N.sub.min=1 coinciding with .sub.BL/.sub.m, in (6). Similar to .sub.BL/.sub.m, in (6), .sub.TL/.sub.m in (5) contains a “Low” SNR region 1001 and a “High” SNR region 1003. Unlike .sub.BL/.sub.m, in (6), which contains a low SNR region 1004, and a high SNR region 1005, .sub.TL/.sub.m in (5) also contains a medium SNR region 1002, denoted as “Med. SNR” in FIG. 10a where doubling .sub.TL/.sub.m in (5) requires a fixed multiple increase in

[00031]$\frac{{\Lambda }_k\mathrm{��}}{N_{\min }}\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{\mathrm{��}}_m}$

since its contribution towards .sub.TL is mostly linear.

[0113] Interpretation of FIG. 10a: The medium SNR region 1002 is created in .sub.TL in (5) when the average received BL SNR

[00032]$\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{\mathrm{��}}_m},$

is >1 while

[00033]$\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{\mathrm{��}}_m}\frac{{\Lambda }_k\mathrm{��}}{N_{\min }}$

is «1. In other words,

[00034]$\frac{{\Lambda }_k\mathrm{��}}{N_{\min }}$

must be much smaller than

[00035]$\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{\mathrm{��}}_m}$

in order to create the medium SNR region 1002. The source for having Λ.sub.k small and N.sub.min large while keeping <∞ is having an arbitrarily large number of DOF, while complying with Constraint 2. In a practical design requiring finite latency, all DOF must have a Finite Access Time (FAT), or equivalently, the time it takes to access any such DOF is finite. We refer to such DOF as FAT, and observe that only TL systems have an arbitrarily large number of FAT DOF in their high frequency components, while BL systems have only a finite number of FAT DOF since they are not allowed to contain high frequency components.

[0114] Attribute of h for to be <∞: Given that some of the communication channels 106, 306, 406 in this disclosure are to be constrained by a spectral mask 701 with <∞, it is imperative to analyze .sub.x.sub.k.sub.(t)(f) 801. An important attribute of h 1101, which affects the spectral decay of .sub.x.sub.k.sub.(t)(f) 801, is the Degree of Differenciability (DOD), .sub.k, of the k.sup.th column, {h}.sub.k of h 1101, defined as the number of times {h}.sub.k can be differenced in time until a Dirac delta impulse, δ, appears. Mathematically, this implies that

[00036]$\stackrel{\Delta }{=}\underset{{\left\{{\stackrel{.fwdarw.}{\Delta }}_k^n\right\}}_l=\delta }{\min }\left\{n\right\}$

where {{right arrow over (Δ)}.sub.k.sup.n}.sub.l is the l.sup.th element in the differencing vector, {right arrow over (Δ)}.sub.k.sup.n, of order n, corresponding to {h}.sub.k, and defined as

[00037]${\left\{{\stackrel{.fwdarw.}{\Delta }}_k^n\right\}}_l\stackrel{\Delta }{=}{\left\{{\stackrel{.fwdarw.}{\Delta }}_k^{n-1}\right\}}_{l+1}-{\left\{{\stackrel{.fwdarw.}{\Delta }}_k^{n-1}\right\}}_l\mathrm{for}l\in \left\{1,\mathrm{.Math.},N-n\right\},n\in \mathrm{��}$

with initial condition:

[00038]${\left\{{\stackrel{.fwdarw.}{\Delta }}_k^0\right\}}_l\stackrel{\Delta }{=}{\left\{h\right\}}_{k,l},l\in \left\{1,\mathrm{.Math.},N-1\right\}.$

[0115] Examples of : [0116] When {h}.sub.k is a TL rectangular pulse, .sub.k=1. [0117] When {h}.sub.k is one lobe of a sine wave, .sub.k=2. [0118] When {h}.sub.k is a pseudo-noise (PN) sequence, .sub.k=0.

[0119] The following 2 DOD properties are used below: [0120] DOD Property I: When {h}.sub.k is the sum of two TL vectors, {h.sub.1}.sub.k, and {.sup.2}.sub.k, i.e. {h}.sub.k={h.sub.1}.sub.k+{h.sub.2}.sub.k, with respective DOD, .sub.k,1 and .sub.k,2, its resulting DOD, .sub.k, is asymptotically equal to

[00039]$\underset{N.fwdarw.\infty }{\lim }{k}_{}=\min \left({k,1}_{},{k,2}_{}\right).$ [0121] DOD Property II: When {h}.sub.k is the linear or circular convolution between two TL vectors, {h.sub.1}.sub.k and {h.sub.2}.sub.k with respective DOD, .sub.k,1 and .sub.k,2, its resulting DOD, .sub.k, is: .sub.k=.sub.k,1+.sub.k,2

[0122] Theorem II derives the slope of the medium SNR region 1002 as a function of the DOD of h.sub.Ch.

[0123] Theorem II: Doubling .sub.TL in (4) across its medium SNR region 1002 requires increasing

[00040]$\frac{\stackrel{\_}{\mathrm{��}}K\mathrm{��}}{{\mathbb{N}}_o{f}_s}$

by a fixed multiple of where .sub.k is the DOD of h.sub.Ch.

[0124] The following constraint derives the modulation, which maximizes n, when using a Minimum Mean Square Error with Successive Interference Cancellation (MMSE-SIC) detector 110, 210, 310 at Rx 108, 208, 308, selected for its low complexity and its asymptotic optimality under certain conditions. This constraint maximizes .sub.TL in the medium SNR region 1002. Modulation Constraint: It is possible to show that minimizing the arithmetic mean of the MMSE at Rx 108, 208, 308 is equivalent to maximizing

[00041]$\frac{d}{dSN{R}_k}{\log }_2\left(1+\frac{SN{R}_k}{{\mathrm{��}}_k}\right)\forall k$

where SNR.sub.k is the received normalized SNR corresponding to {h.sub.Ch}.sub.k while

[00042]$\frac{1}{k_{}}$

is its muitiuser efficiency.

[0125] Unlike water-filling, which deals with parallel channels, the solution for such optimization is

[00043]$\frac{{\mathrm{SNR}}_k}{k_{}}=1\forall k.$

This implies that the modulation of choice for the elements of {right arrow over (α)} corresponds to loading each DOF with about 1 bit of information. In comparison, the low SNR regions 1001, 1004 correspond to loading<1 bit/DOF, while the high SNR regions 1003, 1005 typically correspond to loading>1 bit/DOF. When .sub.k≠ for some k, we use instead:

[00044]$\begin{array}{cc}\underset{k}{\min }\left\{{\mathrm{SNR}}_k\right\}\ge \underset{k}{\max }\left\{k_{}\right\}.& \mathrm{Constraint}3\end{array}$

[0126] Theorem III modifies Theorem Ito include a mask constraint 701 and a modulation constraint. Theorem III: The capacity, .sub.TL, of the TL channel corresponding to h.sub.Ch in (3), subject to

[0127] Constraints 1-3, with K≤d and with .sub.k>0 is

[00045]$\begin{array}{cc}{\mathrm{��}}_{\mathrm{TL}}=\frac{{\mathrm{��}}_m}{}\frac{N}{M}\underset{k=1}{\overset{Q}{.Math.}}{\log }_2\left(1+\frac{{\mathrm{SNR}}_k}{k_{}}\right)\mathrm{bps}& \left(7\right)\end{array}$

where .sub.k is asymptotically

[00046]$\propto \frac{1}{}{-1}^{}$

as N»1 with

[00047]$N_{\mathrm{��}}\stackrel{\Delta }{=}.Math.N\text{/}\mathrm{��}.Math.,$

the ceiling of N/, and N≥N.sub.min, using an MMSE-SIC detector 110, 210, 310 at Rx 108, 208, 308.

[0128] Importance of Theorem III: In FIG. 10a.sub.TL/.sub.m in (7) is illustrated as a number of points (shown with “square” markers), each corresponding to a value of N.sub.min. FIG. 10a confirms that Constraint 3 maximizes .sub.TL in the medium SNR region. Based on (7) and on Constraint 3,

[00048]$\underset{L.fwdarw.\infty }{\lim }{\mathrm{��}}_{\mathrm{TL}}\ge \frac{{\mathrm{��}}_m}{\mathrm{��}}\frac{\mathrm{NK}}{\mathrm{��}}.$

Therefore, doubling

[00049]$\underset{L.fwdarw.\infty }{\lim }{\mathrm{��}}_{\mathrm{TL}}$

by doubling

[00050]$\frac{N}{\mathrm{??}},$

for a fixed

[00051]$\frac{{\mathrm{��}}_m}{\mathrm{��}}$

and K, requires increasing SNR.sub.k by a fixed multiple of , while doubling

[00052]$\underset{L.fwdarw.\infty }{\lim }{\mathrm{��}}_{\mathrm{TL}}$

by doubling K, for a fixed

[00053]$\frac{{\mathrm{��}}_m}{\mathrm{��}}\mathrm{and}\frac{N}{\mathrm{��}},$

requires increasing SNR.sub.k by a fixed multiple of when K≤d.

[0129] The next section introduces a novel TL system with FAT DOF, referred to as an MTF system.

5.2 MTF Design

[0130] Design Problem: h 1101 in (1) is to be designed based on Theorems I and III with the goal of achieving a desired channel capacity, 1301, for a given channel 106, 306, 406 of BW. Three design steps, MTF Design Steps I-III 1302, 1304, 1306, are shown below, followed by a proposed MTF design implementation. All 3 steps attempt to design h 1101 such that the minimum required average received SNR is minimized for a given desired capacity, 1301, and for a given BW . This requires designing h 1101 s.t. the set, {Λ.sub.k}.sub.k=1.sup.rank (h.sup.CH.sup.), of squared singular values of h.sub.Ch in (3) has a variance that is minimized while complying with Constraint 2.

[0131] MTF Design in Solution:

[0132] First, we define

[00054]$T_{s,\max }\stackrel{\Delta }{=}\frac{K}{{\mathrm{����}}_{\mathrm{��}}}\ge 1\text{/}{f}_s$

1301 as a function of K and , which depend on the selected TL channel. For example, when the TL channel has relatively low interference, such as with K=1, one can select the TL system to be with memory, i.e. with =1<N, implying that

[00055]$T_{s,\max }=\frac{1}{{\mathrm{��}}_{\mathrm{��}}}\ge 1\text{/}{f}_s\mathrm{or}{f}_s\ge {\mathrm{��}}_{\mathrm{��}}$

1301. On the other hand, when the TL channel has relatively high interference, i.e. with K»1, one can select the TL system to be memoryless, i.e. with =N, implying that

[00056]$T_{s,\max }=\frac{K}{N{\mathrm{��}}_{\mathrm{��}}}\ge 1\text{/}{f}_s\mathrm{or}{f}_s\ge \frac{N}{K}{\mathrm{��}}_{\mathrm{��}}$

1301.

[0133] MTF Design Step I 1302: For a fixed T.sub.s≤T.sub.s,max 1301, select the number, N 1303, of FAT DOF as

[00057]$N\stackrel{\Delta }{=}{N}_{\mathrm{��}}+N_{\mathrm{��}},$

where: [0134] a) is defined as the number of Shaping FAT (S-FAT) DOF 1303 selected such that ≥N.sub.min in order to comply with the BW constraint, .sub.TL≤.sub.m, of .sub.Mask(f) 701; and [0135] b) is defined as the number of Interpolating FAT (I-FAT) DOF 1303 obtained through the creation of interpolated sampled frequencies 1303 inside the existing occupied band 705. [0136] Power is taken from existing frequencies and allocated to the newly formed frequencies 1303 s.t. Constraint 1 is preserved. Commonly, =0 since no constraints depend on .

[0137] MTF Design Step II 1304: Once N 1303 is selected and the newly sampled frequencies are created 1303, .sub.k can be reduced by equalizing the power, E{|H.sub.k(Ω)|.sup.2}, across Ω∈[−π, π] as much as possible, while preserving Constraint 2, where H.sub.k(Ω) is the Discrete-Time Fourier Transform (DTFT) of the k.sup.th column, {h.sub.Basic}.sub.k, of h.sub.Basic 1102, 1103 and Ω is the normalized frequency. This equalization is defined as taking power from frequency samples 1303 with above average power and allocating it to frequencies 1305 with below average power, thereby preserving Constraint 1. It can be shown using Karamata's inequality that such power allocation reduces the variance of {Λ.sub.k}.sub.k=1.sup.rank (h.sup.Ch.sup.) hence increasing .sub.TL.

[0138] MTF Design Step III 1306: Once N 1303 is selected, the newly sampled frequencies 1303 are created, and the power, E{|H.sub.k(Ω)|.sup.2}, across Ω∈[−π, π] is equalized 1305 as much as possible, .sub.k can be reduced by selecting the phases of the samples of H.sub.k(Ω), s.t. the entries 1307 of h.sub.Basic 1102, 1103 are zero mean RVs, (ideally) Gaussian. This assignment of the phases in H.sub.k (Ω) does not affect the power, E{|H.sub.k(Ω)|.sup.2}, across Ω∈[−π, π], and thus, preserves Constraints 1-2.

[0139] Nomenclature: We refer to h 1101 designed based on MTF Design Steps I-III 1302, 1304, 1306, and subject to Constraints 1-3, as an MTF matrix. In this case, we denote h 1101 as h.sub.MTF 1104, h.sub.Ch as h.sub.MTF,Ch, h.sub.Basic 1102, 1103 as h.sub.B_MTF, .sub.TL in (7) as .sub.MTF and refer to the combination of the MTF system and of the channel as the MTF channel. h.sub.B_MTF is defined by the building blocks h.sub.B_MTF,1 1105, . . . , , 1106. The k.sup.th column {h.sub.B_MTF}.sub.k 1308 of h.sub.B_MTF is obtained from H.sub.k(Ω) using an inverse DTFT 1310.

MTF Design Implementation: An implementation of MTF Design Steps I-III 1302, 1304, 1306 is proposed here where the k.sup.th column, {h.sub.B_MTF}.sub.k 1308, of h.sub.B_MTF, is expressed as a sum:

[00058]$\begin{array}{cc}{\left\{h_{B\_\mathrm{MTF}}\right\}}_k\stackrel{\Delta }{=}+& \left(8\right)\end{array}$

of 2 vectors, 1208 and 1209, defined as follows:

[0140] Vector I: ∈.sup.N×1 1208 is a pulse vector with a DOD, .sub.k>0, selected in order for {h.sub.MTF}.sub.k to comply with the BW constraint, i.e. .sub.TL≤.sub.m 705. It is formed using .sub.k linear convolutions (each denoted by ‘*’ 1207):

[00059]$\begin{array}{cc}\stackrel{\Delta }{=}\left({\stackrel{.fwdarw.}{h}}_{1,k}.Math.{\stackrel{.fwdarw.}{g}}_{1,k}\right)*\mathrm{.Math.}*\left(.Math.\right)*{\stackrel{.fwdarw.}{b}}_k& \left(9\right)\end{array}$

between .sub.k+1>0 vectors, with the l.sup.th vector, ({right arrow over (h)}.sub.l,k{right arrow over (g)}.sub.l, k) 1215, 1216 for l≤.sub.k, formed as a circular convolution (denoted by ‘’1202, 1205) between a zero mean pseudo-random (PR) vector, {right arrow over (g)}.sub.l,k∈.sup.N.sup.l.sup.×1 1203, 1206 with a DOD=0, and a vector pulse, {right arrow over (h)}.sub.l,k∈.sup.N.sup.l.sup.×1 1201 1204 with a DOD=1; while {right arrow over (b)}.sub.k∈ 1215 is a zero mean PR vector with a DOD=0. The first .sub.k−1 linear convolutions produce

[00060]$N_{\mathrm{��}}\stackrel{\Delta }{=}{N}_l-k_{}+1S\text{-}\mathrm{FAT}\mathrm{DOF},$

while the last produces

[00061]$N_{\mathrm{��}}\stackrel{\Delta }{=}{N}_{\mathrm{��}+1}-1I\text{-}\mathrm{FAT}\mathrm{DOF}.$

[0141] Vector II: .sub.k,0∈.sup.N.sup.0.sup.×1 1209 is a PR vector with a DOD=0 selected s.t. {h.sub.MTF}.sub.k complies with the FOSE 707 constraint of .sub.Mask(f) 701, i.e. with a power level

[00062]${\mathrm{��}}_{x_k\left(t\right)}\left(f\right)\le \frac{1}{ℛ}$

in the FOSE band 707. It is possible to generalize N.sub.0 so that it is not necessarily equal to N. For example, it is possible to select N.sub.0=0, implying that 1209 is not included in (8), or equivalently

[00063]${\left\{h_{B\_\mathrm{MTF}}\right\}}_k\stackrel{\Delta }{=}.$

It is also possible to select N.sub.0>N. In this case, N.sub.0−N zeros must be appended to 1208 in (9) in order for 1208 and {h.sub.B_MTF}.sub.k to have a total length of N.sub.0.

[0142] The reasoning behind separating {h.sub.B_MTF}.sub.k in (8) into two vectors, 1208 and 1209, is that it is difficult to simultaneously comply with the BW constraint, i.e. with .sub.TL<.sub.m 705; and the FOSE 707 constraint, i.e. with

[00064]${\mathrm{��}}_{x_k\left(t\right)}\left(f\right)\le \frac{1}{ℛ}$

using a single vector with a single DOD.

[0143] By taking advantage of DOD Property I, summing 1208 and 1209 results in {h.sub.MTF}.sub.k having a DOD .sub.k, since 1209 has a DOD 0.

[0144] The reasoning behind using .sub.k−1 circular convolutions 1202, 1205 in 1208 in (8) is that it is difficult to use a single vector with a single DOD while achieving the following 2 requirements: (1) the entries of {h.sub.B_MTF}.sub.k are zero mean RVs; while (2) {h.sub.MTF}.sub.k complies with the BW constraint that .sub.TL≤.sub.m 705. By taking advantage of DOD Property II, circularly convolving {right arrow over (h)}.sub.l,k 1201, 1204 with {right arrow over (g)}.sub.l,k 1203, 1206, produces a vector with a DOD=1 since the DOD for {right arrow over (g)}.sub.l,k 1203, 1206 is 0, implying that 1208 has a DOD .sub.k. The pulse, 1208, is made to comply with .sub.TL≤.sub.m 705 by properly selecting .sub.k and .

[0145] Theorem IV: The MTF channel corresponding to {h.sub.B_MTF}.sub.kin (8) with =0 under Constraints 1-3, has a capacity, .sub.MTF, identical to .sub.TL in (7) except .sub.k is proportional to

[00065]$\begin{array}{cc}{\mathrm{��}}_k\propto \left(\frac{1+\frac{N_0}{\mathrm{\lambda ℛ}}}{\frac{1}{\frac{1}{\mathrm{��}}}+c_{N_0,0}\frac{\mathrm{��}{N}_0}{\mathrm{\lambda ℛ}N}}\right)\mathrm{as}N\text{>>}1& \left(10\right)\end{array}$

where and c.sub.N.sub.0.sub.,0 are the coefficients of proportionality corresponding to and in (8) respectively, with

[00066]$\lambda \stackrel{\Delta }{=}\mathrm{��}\text{/},$

using an MMSE-SIC detector 110, 210, 310 at Rx 108, 208, 308.

[0146] Importance of Theorem IV: Based on (10), .sub.MTF consists of two medium SNR regions 1006, 1007, as shown in FIG. 10b. As N increases, doubling .sub.MTF initially requires increasing the SNR by a fixed multiple of , which is the first medium SNR region 1006, referred to as “1.sup.st Med. SNR” in FIG. 10b. Then, later on it requires increasing the average received SNR by a fixed multiple of 2, which is the second medium SNR region 1007, referred to as “2.sup.nd Med. SNR” in FIG. 10a. FIG. 10a compares .sub.MTF/.sub.m based on (10) with .sub.BL/.sub.m in (6), where =.sub.BL=1, =1, =K=1 and in (8) is a rectangular pulse with in (8) 30 dBr below . In FIG. 10b, .sub.MTF/.sub.m without Constraint 3 is illustrated as a number of curves (with “.” markers), each curve corresponding to a value of N.sub.min.Math..sub.MTF/.sub.m with Constraint 3 is illustrated as a number of points (with “square” markers), each point corresponding to a value of N.sub.min. FIG. 10b confirms that indeed Constraint 3 maximizes .sub.MTF/ in both medium SNR regions 1006, 1007.

[0147] Under certain conditions, the following asymptotic limits can be reached: [0148] a) When

[00067]$\frac{N_0}{\mathrm{\lambda ℛ}}\text{<<}1,$

Theorem IV reduces to Theorem III. [0149] b) When

[00068]$\frac{N_0}{\mathrm{\lambda ℛ}}\text{>>}1\mathrm{and}{c}_{N_0,0}\frac{\mathrm{��}{N}_0}{\mathrm{\lambda ℛ}N}\text{>>}\frac{1}{c_{K,N,d,\mathrm{��}}\frac{1}{\mathrm{��}}{K}^{2{\mathrm{��}}_k-1}{N}_{\mathrm{��}}^{2{\mathrm{��}}_k}},$

we have

[00069]$\begin{array}{cc}\underset{N_0.fwdarw.\infty }{\lim }{\mathrm{��}}_k=\frac{N}{\mathrm{��}}& \left(11\right)\end{array}$

[0150] This limit applies to the case when .sub.Mask(f) 701 corresponds to the IEEE 802.11 WLAN mask, denoted as .sub.WiFi(f). [0151] c) When

[00070]$\frac{N_0}{\mathrm{\lambda ℛ}}\stackrel{\Delta }{=}\mathrm{��}\text{<<}1$

with q, a constant and

[00071]$c_{N_0,0}\frac{\mathrm{��}{N}_0}{\mathrm{\lambda ℛ}N}\text{>>}\frac{1}{c_{K,N,d,{\mathrm{��}}_k}\frac{1}{\mathrm{��}}{K}^{2{\mathrm{��}}_k-1}{N}_{\mathrm{��}}^{2{\mathrm{��}}_k}},$

we have

[00072]$\begin{array}{cc}\underset{N_0.fwdarw.\infty }{\lim }{\mathrm{��}}_k=\frac{N}{\mathrm{����}}& \left(12\right)\end{array}$

This limit applies to the case when .sub.Mask(f) 701 corresponds to the 3GPP LTE (E-UTRA) mask, denoted as .sub.LTE(f).

5.3 MTF Architecture

[0152] Section 5.3.1 introduces the constraints that are generally imposed on communication systems such as standard-imposed spectral masks 701, as well as the effects of fading and interference across the communication channel 106, 306, 406. Section 5.3.2 proposes several MTF designs based on the constraints introduced in Section 5.3.1, while Section 5.3.3 introduces an architecture that is suitable for allowing various MTF systems to communicate with each other when co-located while using the same band.

5.3.1 Design Constraints

[0153] First, we select 2 important .sub.Mask(f) 701, namely .sub.WiFi(f) and .sub.LTE(f). Then, we model the communication channel 106, 306, 406 and examine its effects on the MTF architecture including the types of interference and restricted bands across such a channel.

[0154] Selection of .sub.Mask(f) 107: In order to include Constraint 2 in Design Steps I-III 1302, 1304, 1306, and in order to derive a fair comparison with some of the existing systems, we define .sub.WiFi(f) and .sub.LTE(f): [0155] a) The 3GPP LTE (E-UTRA) mask, .sub.LTE(f), is defined for a 1.4, 3, 5, 10, 15 and 20 MHz BW, as having≤1% OOBE BW, or equivalently, must contain≥99% of the total integrated mean power in x.sub.k(t) 105, 205, 305 ∀k. [0156] b) The IEEE 802.11 WLAN mask for a 20 MHz BW is:

[00073]${\mathbb{P}}_{\mathrm{WiFi}}\left(f\right)\stackrel{\Delta }{=}\{\begin{array}{cc}0\mathrm{dBr}& .Math.f.Math.\le 9\mathrm{MHz}\\ \mathrm{line}\mathrm{from}0\mathrm{dBr}\mathrm{to}-20\mathrm{dBr}& 9\mathrm{MHz}\le .Math.f.Math.\le 11\mathrm{MHz}\\ \mathrm{line}\mathrm{from}-20\mathrm{dBr}\mathrm{to}-28\mathrm{dBr}& 11\mathrm{MHz}\le .Math.f.Math.\le 20\mathrm{MHz}\\ \mathrm{line}\mathrm{from}-28\mathrm{dBr}\mathrm{to}-40\mathrm{dBr}& 20\mathrm{MHz}\le .Math.f.Math.\le 30\mathrm{MHz}\\ -40\mathrm{dBr}& 30\mathrm{MHz}\le .Math.f.Math.\end{array}$

[0157] In .sub.WiFi(f), the first frequency band, |f|≤9 MHz, corresponds to the occupied band 705 with a bandwidth, =18 MHz. The middle three frequency bands correspond to DOBE 706 with a bandwidth, .sub.OOBE=42 MHz . The last frequency band, If |f|≥30 MHz, correspond to the FOSE band 707 with an infinite bandwidth and a power level 1/=−40 dBr.

[0158] Modeling of the Communication Channel 106, 306, 406: When

[00074]$f_s\stackrel{\Delta }{=}1\text{/}{T}_s\mathrm{is}\ge 8\mathrm{MHz}$

and NT.sub.s≤1 ms, the channel 106, 306, 406 can be modeled as a frequency-selective (FS) slowly fading channel affected by a frequency-dependent path loss (PL) modeled after Friis free-space PL (FSPL) model. Mathematically, such a channel can be modeled as Linear Time-Invariant (LTI) and characterized using a discrete-time random impulse response, {right arrow over (h)}.sub.Ch, of finite length, δN, referred to as the discrete delay spread of the channel. The fading can be modeled either as Rayleigh for a non-LOS channel or as Rician with a strong LOS component for a LOS channel.

[0159] Effects of the Selected Channel model: [0160] 1) Mathematically, the main effect of the frequency-selective channel is to linearly convolve each column, {h.sub.MTF}.sub.k, in h.sub.MTF 1104 with {right arrow over (h)}.sub.Ch. The outcome of such a convolution is a new MTF matrix, h.sub.MTF,Ch∈, defined as

[00075]$\begin{array}{cc}{h}_{\mathrm{MTF},\mathrm{Ch}}\stackrel{\Delta }{=}{\mathrm{toep}}_d\left\{{\stackrel{.fwdarw.}{h}}_{\mathrm{MTF},\mathrm{Ch}}\right\}& \left(13\right)\end{array}$

where h.sub.MTF,Ch∈has h.sub.B_MTF replaced by h.sub.B_MTF,Ch∈with M replaced by

[00076]$ℳ\stackrel{\Delta }{=}\mathrm{��}+\mathrm{��}\left(L-1\right),$

N replaced by

[00077]$\mathrm{��}\stackrel{\Delta }{=}N+\delta N-1$

and N.sub.o replaced by

[00078]${\mathrm{��}}_o\stackrel{\Delta }{=}{N}_o+\delta N-1.$

The increase in N and in N.sub.o by δN−1 is equivalent to an increase in the number, , of I-FAT DOF in the MTF system by δN. Based on DOD Property II, the linear convolution between {right arrow over (h)}.sub.Ch and the k.sup.th column, {h.sub.MTF}.sub.k, in h.sub.MTF 1104 implies that the resulting DOD is equal to the sum between the original DOD, .sub.k, and the DOD, .sub.Ch, of the communication channel.

[0161] Based on the adopted frequency-selective fading model of the communication channel 106, 306, 406, .sub.Ch=0. In other words, the resulting DOD of h.sub.MTF,Ch, is equal to the original DOD of h.sub.MTF 1104 when the communication channel 106, 306, 406 is FS. [0162] 2) The communication channel 106, 306, 406 can be equivalently characterized in the frequency domain by the DTFT of {right arrow over (h)}.sub.Ch, also referred to as its Transfer Function (TF), H.sub.Ch(Ω). This implies that the following continuous-frequency product:

[00079]$\begin{array}{cc}{H}_{{\mathrm{MTF}}_k,\mathrm{Ch}}\left(\Omega \right)\stackrel{\Delta }{=}{H}_{{\mathrm{MTF}}_k}\left(\Omega \right)H_{\mathrm{Ch}}\left(\Omega \right)\forall \Omega \in \left[-\pi ,\pi \right]& \left(14\right)\end{array}$

can replace the linear discrete-time convolution between {h.sub.MTF}.sub.k and {right arrow over (h)}.sub.Ch where H.sub.MTF.sub.k.sub.,Ch(Ω) is the DTFT of the k.sup.th column, {h.sub.MTF,Ch}.sub.k, of h.sub.MTF,Ch. [0163] 3) Friis FSPL model is based on E{|H.sub.Ch(Ω)|.sup.2} being inversely proportional to |Ω|.sup.2, i.e.

[00080]$\begin{array}{cc}E\left\{{.Math.H_{\mathrm{CH}}\left(\Omega \right).Math.}^2\right\}\propto \frac{1}{{.Math.\Omega .Math.}^2}\forall \Omega \in \left[-\pi ,0\right).Math.\left(0,\pi \right]& \left(15\right)\end{array}$

where E{.} denotes expectation w.r.t. H.sub.Ch(Ω) at Ω, assuming it is ergodic. Based on (15), it is possible to see that the FSPL has a DOD equal to 1. In other words, the effect of the FSPL is to increase the original DOD, .sub.k, of {h.sub.MTF}.sub.k by 1 if the carrier frequency, f.sub.c=0, otherwise, the effect of the FSPL on .sub.k depends on f.sub.c.

[0164] Based on all effects of the communication channel 106, 306, 406, Theorem III is still valid after replacing H.sub.MTF.sub.k(Ω) by H.sub.MTF.sub.k.sub.,Ch(Ω), M by and after re-evaluating .sub.k based on f.sub.c. In order to preserve the original DOD, .sub.k, of h.sub.MTF 1104 a pre-channel filter 500 is required at Tx 104, 204, 304, which is discussed in Section 5.3.2.

[0165] Modeling of Interference: Two types of interference exist across a communication network: [0166] (a) Narrow-Band Interference (NBI), defined as having a width≤125 MHz; and [0167] (b) Wide-Band Interference (WBI), defined as having a width>125 MHz.

[0168] NBI encompasses transmissions from existing systems such as LTE and Wi-Fi systems, and from other MTF systems due to the presence of 1208, while WBI encompasses transmissions from Ultra-Wide Band (UWB) systems and from other MTF systems due to the presence of 1209. Several studies have indicated low utilization of the frequency bands at frequencies>2 GHz as seen in Table I, which displays the average duty cycle versus frequency range≤7,075 MHz based on results in in an urban environment. Table I is consistent with several other studies of urban centers across North America and Europe. All studies indicate an exponential decline in utilization directly proportional to frequency, f. We refer to frequency ranges with known Heavy Utilization as .sub.HU.

TABLE-US-00001 TABLE I Average utilization duty cycle across a frequency range. Frequency range Average utilization duty cycle  75-1000 42.00%  1000-2000 13.30%  2000-3000 3.73% 3000-4000 4.01% 4000-5000 1.63% 5000-6000 1.98% 6000-7075 1.78%

[0169] Restricted Bands, .sub.RB: Further to having to contend with both NBI and WBI, some bands, referred to as .sub.RB, have been deemed restricted by the regulatory bodies (47 CFR 15.205).

5.3.2 Pulse and Filter Design

[0170] Based on the knowledge of the statistics of the communication channel 106, 306, 406 including its model, the types of interference across it and the existence of .sub.RB, the disclosure designs pulses such as {right arrow over (h)}.sub.l,k 1201, 1204, 1209, {right arrow over (g)}.sub.l,k 1203, 1206 as well as filters such as a Pre-channel filter 500 at Tx 104, 204, 304, and a Post-channel filter 615 at Rx 108, 208, 308 with the goal of optimizing .sub.MTF subject to Constraints 1-3.

[0171] Design of {right arrow over (h)}.sub.l,k∈.sup.N.sup.l.sup.×1 1201, 1204: A fundamental design for {right arrow over (h)}.sub.l,k 1201, 1204 in (9) is a rectangular pulse. Even though it has all its zeros on the unit circle, it is possible to move its zeros away from the unit circle by shifting it in frequency by 1/(2N.sub.lT.sub.s). This shift forms a complex pulse, {right arrow over (h)}.sub.REC,l,k 1212, 1213, with a real part in the shape of one lobe of a cosine wave and an imaginary part in the shape of a negative lobe of a sine wave. When all {right arrow over (h)}.sub.l,k ∀ 1201, 1204 are selected as {right arrow over (h)}.sub.REC,l,k 1212, 1213, 1208 in (9) is denoted as 1214. When 1208 in (9) is selected as

[00081]${\mathrm{REC},1}_{},\lambda \stackrel{\Delta }{=}$

in Theorem III is asymptotically equal to π. In this case, the amplitude of .sub.REC,1 is selected to comply with Constraint 1.

[0172] Design of .sub.k,0 ∈.sup.N.sup.0.sup.×1 1209: A possible design of .sub.k,0 1209 in (8) is:

[00082]$\begin{array}{cc}{k,0}_{}=\left\{{\left[e^{j{\vartheta }_{k,1}}\mathrm{.Math.}{e}^{j{\vartheta }_{k,N_0}}\right]}^T\right\}& \left(16\right)\end{array}$

where denotes an inverse Discrete Fourier Transform (DFT) operation; and the phase, .sub.k,i∈{0,2π}, is chosen as PR with a uniform distribution across {0,2π} for 1≤i≤N.sub.0. .sub.k,0 1209 is also known as a frequency-based PR polyphase signature.

[0173] Design of {right arrow over (g)}.sub.l,k∈.sup.N.sup.l.sub.×1 1203, 1206: A possible design of {right arrow over (g)}.sub.l,k 1203, 1206 in (9) is

[00083]$\begin{array}{cc}{\stackrel{.fwdarw.}{g}}_{l,k}={\mathrm{��ℱ��}}^{-1}\left\{{\left[e^{j{\vartheta }_{l,k,1}}\mathrm{.Math.}{e}^{j{\vartheta }_{l,k,N_l}}\right]}^T\right\}& \left(17\right)\end{array}$

with the phase, .sub.l,k,i∈{0,2π}, chosen as PR with a uniform distribution across {0,2π} for 1≤i≤N.sub.l, similar to .sub.k,0 1209 in (16), except that .sub.l,k,1 must equal .sub.l,k, N.sub.l in order to minimize the overhead factor, . Since the communication channel forces H.sub.MTF.sub.k(Ω) in (14) to be multiplied by H.sub.Ch(Ω), the resulting product, H.sub.MTF.sub.k.sub.,Ch(Ω), forces {right arrow over (g)}.sub.l,k 1203, 1206 in (17) to be replaced by

[00084]$\begin{array}{cc}{\stackrel{.fwdarw.}{g}}_{l,k,\mathrm{Ch}}={\mathrm{��ℱ��}}^{-1}\left\{{\left[.Math.a_{l,k,1}.Math.e^{j{\vartheta }_{l,k,1}}\mathrm{.Math.}.Math.a_{l,k,{\mathrm{��}}_l}.Math.e^{j{\vartheta }_{l,k,{\mathrm{��}}_l}}\right]}^T\right\}& \left(18\right)\end{array}$

when =1 and .sub.k=1 where |α.sub.l,k,1|, . . . , || are random amplitudes, which have either a Rician distribution with a strong LOS component across a LOS channel, or a Rayleigh distribution across a NLOS channel, and

[00085]${\mathrm{��}}_l\stackrel{\Delta }{=}{N}_l+\delta N-1.$

[0174] Selection of when .sub.Mask(f)=.sub.LTE(f): The disclosure selects in (12) as 0.5% and allocates the remaining 0.5% to the DOBE 706 BW in 1208. Under Constraint 2, (12) can be re-written as

[00086]$\underset{N_0.fwdarw.\infty }{\lim }{\mathrm{��}}_k={\left(N\text{/}\mathrm{��}\right)}_{\mathrm{dB}}+23\mathrm{dB}.$

[0175] Design of Pre-channel filtering 500 at Tx 104, 204, 304: In order to comply with .sub.RB (47 CFR 15.205), and to prevent transmitting across .sub.HU, a pre-channel filter 500 is recommended at Tx 104, 204, 304. Furthermore, according to (15), the effect of the FSPL is to increase the DOD, .sub.k, of h.sub.MTF 1104 by 1 despite the fact that .sub.k,0 1209 has been added in (8) to force the resulting DOD to asymptotically take the value 0. In order to address all 3 concerns, h.sub.MTF 1104 is replaced by a pre-channel MTF matrix, , based on replacing in (8) 1208 by , .sub.k,0 1209 by and {h.sub.MTF}.sub.k by

[00087]${\left\{h_{\mathrm{MTF}}^{\mathrm{��}}\right\}}_k\stackrel{\Delta }{=}{\stackrel{.fwdarw.}{\mathrm{��}}}_{k,\mathrm{��}}^{\mathrm{��}}+{\stackrel{.fwdarw.}{\mathrm{��}}}_{k,0}^{\mathrm{��}}$

with a DTFT (Ω) pre-processed by the following 2 actions: [0176] 1. Pre-distort (Ω) by |Ω| by ∀Ω∈[−π, 0) ∪(0, π]as

[00088]$\begin{array}{cc}{H}_{{\mathrm{MTF}}_k}^{\mathrm{��}}\left(\Omega \right)=\frac{.Math.\Omega .Math.}{\sqrt{v}}{H}_{{\mathrm{MTF}}_k}\left(\Omega \right)\forall \Omega \in \left[-\pi ,0\right).Math.\left(0,\pi \right]& \left(19\right)\end{array}$

where v is selected to keep

[00089]$E\left\{{.Math.H_{MT{F}_k}^{\mathrm{��}}\left(\Omega \right).Math.}^2\right\}\le \frac{1}{ℛ}\forall \Omega \in \left[-\pi ,0\right).Math.\left(0,\pi \right].$ [0177] 2. Force (Ω) to contain a null at the excluded band:

[00090]${ℬ}_{\mathrm{ex}}\stackrel{\Delta }{=}\left\{\left\{{ℬ}_{\mathrm{RB}}.Math.{ℬ}_{\mathrm{HU}}\right\}.Math.{ℬ}_{\mathrm{TL}}\right\}.Math.{\stackrel{\_}{ℬ}}_{\mathrm{TL}}$

with

[00091]${ℬ}_{HU}\stackrel{\Delta }{=}\left[0,2\mathrm{GHz}\right]$

and .sub.TL the complement of

[00092]${ℬ}_{\mathrm{TL}}\stackrel{\Delta }{=}\left[f_c-f_s/4,f_c+f_s/4\right]\mathrm{and}{f}_c\ge f_s/4.$

[0178] As a result of both actions, .sub.k in (10) is replaced by

[00093]$\begin{array}{cc}{\mathrm{��}}_k\propto \left(\frac{1+\frac{N_0}{\lambda vℛ}}{c_{K,N,d,\aleph }\frac{1}{\mathrm{��}}{K}^{2\aleph -1}{N}_d^{2\aleph }+c_{N_0,0}\frac{\mathrm{��}{N}_0}{\lambda vℛN}}\right)\mathrm{as}N⪢1& \left(20\right)\end{array}$

and (11) is replaced by

[00094]$\underset{N_0.fwdarw.\infty }{\lim }{\mathrm{��}}_k.fwdarw.{\left(\frac{N}{\mathrm{��}}\right)}_{dB}-5.16\mathrm{dB}$

since v=2(π.sup.2/6) according to Basel problem where

[00095]$2\left(\underset{N_0.fwdarw.\infty }{\lim }{.Math.}_{n=1}^{N_0}\frac{1}{n^2}\right)=2\left({\pi }^2/6\right)=5.16\mathrm{dB}$

when the samples are real. On the other hand, (12) is replaced by

[00096]$\underset{N_0.fwdarw.\infty }{\lim }{\mathrm{��}}_k={\left(\frac{N}{\mathrm{��}}\right)}_{dB}-{\left(\mathrm{��}\right)}_{dB}-5.16\mathrm{dB}.$

For example, when

[00097]$\mathrm{��}=0.5\%,\underset{N_0.fwdarw.\infty }{\lim }{\mathrm{��}}_k={\left(\frac{N}{\mathrm{��}}\right)}_{\mathrm{dB}}+17.9\mathrm{dB}.$

[0179] Design of Post-channel filtering 615 at Rx 108, 208, 308: Post-channel filtering 615 can be used at Rx 108, 208, 308 to reduce the effects of NBI across the communication channel. In this case, it must include an MTF excision filter, which consists of the following two steps: [0180] (a) Estimate the frequency range, .sub.NBI, corresponding to NBI. A frequency, f, belongs to .sub.NBI when .sub.y(t)(f)≥.sub.x.sub.k.sub.(t)(f)+.sub.Th where .sub.Th is a threshold selected to meet a certain optimization criterion for reducing NBI. [0181] (b) Excise the estimated NBI by forcing a null in the PSD, .sub.y(t)(f), of the continuous-time version, y(t), of {right arrow over (r)}, at f∈.sub.NBI.

[0182] Post-channel filtering should also include a null at

[00098]${ℬ}_{\mathrm{NBI}}.Math.{ℬ}_{\mathrm{ex}}\stackrel{\Delta }{=}\left\{\left\{{ℬ}_{\mathrm{NBI}}.Math.{ℬ}_{\mathrm{RB}}.Math.{ℬ}_{\mathrm{HU}}\right\}.Math.{ℬ}_{\mathrm{TL}}\right\}.Math.{\stackrel{\_}{ℬ}}_{\mathrm{TL}}$

in order to reduce the effect of noise and interference at Rx 108, 208, 308.

[0183] Selection of Sampling Type and frequency, f.sub.s: There are 3 types of sampling available in communication systems: baseband sampling, IF sampling and RF sampling. RF sampling is recommended when f.sub.s≥4f.sub.c, since it does not require any up-conversion/down-conversion stages as shown in FIGS. 6a and 6b. In this case, the Carrier Frequency, f.sub.c, is selected≤f.sub.s/4. On the other hand, IF sampling is recommended over baseband sampling when f.sub.s<4f, since it requires a smaller number of conversion stages as shown in FIGS. 5a and 5b. In this case, the Intermediate Frequency, f.sub.IF, is selected equal to f.sub.s/4.

[0184] Selection of Carrier Frequency f.sub.c: In order to select a frequency range s.t. f.sub.c≤f.sub.c/4 with relatively low interference and low path loss, while allowing for a multipath-rich environment that is suitable for MIMO communications and while avoiding .sub.HU, the disclosure proposes to select f.sub.c∈[2 GHz, 6 GHz]. It is possible to reduce .sub.k ∀k by selecting f.sub.c for the k.sup.th Tx to be distinct from all the other K−1 carrier frequencies ∀k. An optimal selection of the K carrier frequencies is for each frequency to select one unique frequency either from the optimal set

[00099]$S_{f_c}\stackrel{\Delta }{=}\left\{\frac{f_s}{4K},\frac{3f_s}{4K},\mathrm{.Math.},\frac{\left(2K-1\right)f_s}{4K}\right\}$

or from any other set such as

[00100]$S_{f_c}^{\prime }\stackrel{\Delta }{=}\left\{\frac{\left(K+1\right)f_s}{8K},\frac{\left(K+3\right)f_s}{8K},\mathrm{.Math.},\frac{\left(3K-1\right)f_s}{8K}\right\}.$

[0185] When K>1 and the K carrier frequencies are selected uniquely from S.sub.fc or from S.sub.f.sub.c′, Post-channel filtering 615 containing an MTF excision filter is recommended at Rx 108, 208, 308 in order to reduce NBI, which originates from the other K−1 interfering carrier frequencies.

5.3.3 MTF System Architecture

[0186] The architecture shown in FIGS. 1-4 together with Tx 104, 204, 304 and Rx 108, 208, 308 as described in FIGS. 5a, 5b, 6a, 6b enable an MTF system to communicate with other co-located MTF systems, when using overlapping licensed or unlicensed bands. Such MTF systems include cellular-type, Wi-Fi-type and wireless sensor/Internet-of-Things-type systems, all capable of communicating with one another. FIGS. 1-4 together with FIGS. 5a, 5b, 6a, 6b, have a digital component 510, which includes MTF Modulator 102, 202, 302, a portion of Tx 104, 204, 304, a portion of Rx 108, 208, 308 and MTF detector 110, 210, 310. The digital portion of Tx 104, 204, 304 includes the Pre-Channel filter 500 and the digital portion of Digital-to-Analog (D/A) converter 503. The digital portion of Rx 108, 208, 308 includes the Post-Channel filter 615 and the digital portion of Analog-to-Digital (A/D) converter 613. The digital component is mostly software-defined (S/W) and is allowed to change personality according to the MTF system it is communicating with, by adjusting {h.sub.MTF}.sub.k, Pre-channel filter 500, Post-channel filter 615 and MMSE-SIC detector 110, 210, 310. Also, the architecture shown in FIGS. 1-4 together with a portion of Tx 104, 204, 304 and a portion of Rx 108, 208, 308 have analog components 511, 513, 616, 618, which are hardware-defined (H/W). As described in FIGS. 5a, 5b, 6a, 6b, the analog portions of Tx 104, 204, 304 and Rx 108, 208, 308 include converters (e.g. D/A 502, Up/Down 504, 611 and A/D 613), analog filters 607 (e.g. Band-Pass Filter (BPF) and Low-Pass Filter (LPF)) and amplifiers (e.g. Power Amplifier (PA) 506 and Low Noise Amplifier (LNA) 609).

[0187] Given that most existing systems contain a S/W component and a H/W component, it is possible to upgrade such systems to an MTF system through a S/W download, without requiring a H/W modification as long as it is possible to overcome its limitation. For example, when the sampling frequency, f.sub.A/D, of the available A/D converter is smaller than the required f.sub.s by a multiple >1 s.t. f.sub.A/D=f.sub.s/ , it is possible to reduce f.sub.s by to accommodate f.sub.A/D while maintaining the same desired channel capacity, , using several non mutually exclusive techniques:

[0188] MTF Technique 1: Decimate h.sub.MTF by , while increasing T.sub.s by .

[0189] MTF Technique 2: Relax Constraint 3 by loading >1 bits of information/DOF.

[0190] MTF Technique 3: Select = while forcing each column of to have a distinct f.sub.c.

[0191] It is possible to combine several of the MTF techniques shown above in order to overcome the limitation of having f.sub.A/D=f.sub.s/. For example, by increasing the number of information bits/DOF from 1 bit to 2 bits, while increasing from 1 to 4, we have =8.

5.4 MTF MA Networks

[0192] This section designs MTF MA networks across a centralized topology similar to existing MA networks, such as LTE and Wi-Fi networks. As typical of any centralized topology, the MTF MA network consists of two types of transmissions: (a) downlink (DL) transmissions, from Base Station (BS) or Access Point (AP) to device; and (b) uplink (UL) transmissions, from device to BS/AP. The designs of the MTF MA networks are based on the following assumptions:

5.4.1 Assumptions

[0193] a) Several co-located centralized MTF MA networks use overlapping licensed or unlicensed bands. Based on the system architecture in Section 5.3.3, such networks are capable to collaborate, which offers many benefits. For example Time Division Duplex (TDD) can be implemented, which forces temporal separation between DL and UL transmissions. [0194] b) The MTF BS/AP contains an antenna array to be used in beamforming in the DL portion of most of the MTF MA networks in order to reduce WBI from interfering BSs/APs at devices. [0195] c) The range, .sub.k, between the k.sup.th MTF Tx 104, 304 and an MTF Rx 108, 308 is a function of the link budget, , between them, where

[00101]$\begin{array}{cc}{k,\mathrm{dB}}_{}\stackrel{\Delta }{=}{\mathrm{��}}_{\mathrm{dBm}}-{\mathrm{NF}}_{\mathrm{dB}}-{\left({\mathbb{N}}_o{\mathrm{��}}_m\right)}_{\mathrm{dB}}-\underset{k}{\max }{\left\{{\mathrm{��}}_k^{\mathrm{��}}\right\}}_{\mathrm{dB}}+G_{\mathrm{dB}}& \left(21\right)\end{array}$

[0196] .sub.dBm is the average transmitted power; NF.sub.dB is the Noise Figure of Rx; corresponds to

[00102]$\Delta =f_s\left(1-{\zeta }_{\mathrm{ex}}\right),$

[0197] which is the Noise-equivalent BW in Hz at the output of the post-channel filter; ζ.sub.ex is the excision factor, defined as

[00103]${\zeta }_{\mathrm{ex}}\stackrel{\Delta }{=}\frac{1}{f_s}{\int }_{f\in {ℬ}_{\mathrm{ex}}}\mathrm{fdf};$

G.sub.dB is the antenna gain between Tx and Rx. [0198] d) A unique {right arrow over (g)}.sub.l,k 1203, 1206 as defined in (17) and a unique .sub.k,0 1209 as defined in (16) are selected for the k.sup.th MTF device ∀k. [0199] e) 1208, is designed s.t. its PSD, (f) 808, has nulls at .sub.NBI ∪.sub.ex where (t) is the continuous-time version of 1208. [0200] f) 1209 is designed s.t. its PSD, (f) 809, is pre-distorted according to (19), where (t) is the continuous-time version of 1209. [0201] g) |.sub.UL=N in the UL portion of an MTF MA network in order to decrease

[00104]$\underset{k}{\max }\left\{k_{}\right\}$

in (21), while |.sub.DL=1 in the DL portion in order to maintain a high DL capacity, .sub.MTF, where |.sub.UL and |.sub.DL are the delays in (2) corresponding to UL and DL respectively. An additional reason for selecting |.sub.UL=N is to reduce the Peak-to-Average Power Ratio (PAPR) corresponding to transmissions from an MTF device, which can be reduced even further by selecting 1208 in (9) as .sub.REC,1 1214. A further reason for selecting .sub.UL=N is to have a memoryless MTF MA network with .sub.MTF|.sub.L<∞=.sub.MTF|.sub.L−∞. On the other hand, selecting |.sub.DL−1 in the DL portion implies that the MTF MA network has memory and

[00105]${\mathrm{MTF}}_{}{|}_{L<\infty }=L/\left(L+\frac{N}{d}-1\right){\mathrm{MTF}}_{}.$

For example, when 10N<L<∞, .sub.MTF|.sub.L<∞>0.9.sub.MTF|.sub.L.fwdarw.∞.

[0202] In the UL portion, all Q symbols in {right arrow over (α)} 101, 201, 301, corresponding to all the K active Txs 104, 204, 304, are required to be detected, while in the DL portion, only the desired symbols in a {right arrow over (α)} 101, 201, 301, corresponding to the desired Tx 104, 204, 304, are required to be detected, with the remaining symbols, corresponding to the K.sub.i=K−1 interfering columns in h.sub.MTF 1104, ignored. For this reason, a preferred embodiment is to constrain

[00106]$\underset{k}{\max }\left\{k_{}\right\}$

in (21) to correspond to full implementation of Constraint 3 ∀k for the UL portion, while in the DL portion, a preferred embodiment is to constrain

[00107]$\underset{k}{\max }\left\{k_{}\right\}$

in (21) to correspond to a partial implementation of Constraint 3 corresponding only to the desired received symbols in {right arrow over (α)} 101, 201, 301.

5.4.2 Designs of MTF MA Networks

[0203] Based on the above assumptions, we design 3 MTF MA networks, namely MTF.sub.1, MTF.sub.2 and MTF.sub.3, all constrained by a mask with a BW, .sub.m=20 MHz. This implies that

[00108]${\left({\mathrm{NT}}_s\right)}_{\min }\stackrel{\Delta }{=}/20\mathrm{MHz}$

1301. For example, when .sub.Mask (f) 701 is selected as .sub.WiFi(f) and 1208 as 1214 with a PSD (f) 908, =64/3=21.3 and

[00109]${\left({\mathrm{NT}}_s\right)}_{\min }\stackrel{\Delta }{=}/20\mathrm{MHz}=1.066\mathrm{.Math.s}$

1301. On the other hand, when .sub.Mask(f) 107 is selected as .sub.LTE(f) and 1208 as 1214 with a PSD(f) 908, =40.8 and

[00110]${\left({\mathrm{NT}}_s\right)}_{\min }\stackrel{\Delta }{=}\frac{20\mathrm{MHz}}{}=2.04\mathrm{.Math.s}$

1301.

[0204] Moreover, since the DL portion of each network is assumed to have relatively low interference, it is characterized to be with memory, with =K=1, and

[00111]$T_{s,\max }{|}_{\mathrm{DL}}\stackrel{\Delta }{=}1/{|}_{DL}\ge 1/f_s{|}_{\mathrm{DL}}$

1301. On the other hand, since the UL portion of each network is assumed to have relatively high interference, it is characterized as memoryless, with =N, K»1, and

[00112]$T_{s,\max }{|}_{\mathrm{UL}}\stackrel{\Delta }{=}\frac{K}{N{|}_{UL}}\ge 1/f_s{|}_{\mathrm{UL}}$

1301. Therefore,

[0205] [00113]$N_{\min }{|}_{\mathrm{DL}}\stackrel{\Delta }{=}\frac{{\left({\mathrm{NT}}_s\right)}_{\min }}{T_{s,\max }{|}_{\mathrm{DL}}}=\frac{{|}_{\mathrm{DL}}}{20\mathrm{MHz}}\mathrm{and}$$N_{\min }{|}_{\mathrm{UL}}=\Delta \frac{{\left({\mathrm{NT}}_s\right)}_{\min }}{T_{s,\max }{|}_{\mathrm{UL}}}=\frac{{|}_{\mathrm{UL}}N}{20\mathrm{MHz}K}$

1301.

[0206] Design Parameters for MTF.sub.1, MTF.sub.2 and MTF.sub.3: [0207] 1. MTF.sub.1 is selected to have a desired DL channel capacity of

[00114]${.Math.}_{\mathrm{DL}}=0.2\mathrm{Gbps}$

1301, and a desired UL channel capacity of

[00115]${.Math.}_{\mathrm{UL}}=0.2\mathrm{Gbps}$

1301, both across the unlicensed (Title 47 CFR 15.247) mid-band frequency of f.sub.c.sub.1=2.45 GHz. We also select for both DL and UL,

[00116]$f_{s_1}{.Math.}_{\mathrm{DL}}=f_{s_1}{.Math.}_{\mathrm{UL}}=0.2\mathrm{GHz}$

and IP sampling with

[00117]${ℬ}_{T{L}_1}=\Delta \left[f_{c_1}-f_{s_1}/4,f_{c_1}+\frac{f_{s_1}}{4}\right]\mathrm{and}$${ℬ}_{e{x}_1}\stackrel{\Delta }{=}\left\{\left\{{ℬ}_{{\mathrm{NBI}}_1}.Math.{ℬ}_{RB}.Math.{ℬ}_{HU}\right\}.Math.{ℬ}_{{\mathrm{TL}}_1}\right\}.Math.{\stackrel{\_}{ℬ}}_{{\mathrm{TL}}_1}.$ [0208] 2. MTF.sub.2 is selected to have

[00118]${.Math.}_{\mathrm{DL}}=2\mathrm{Gbps}$

1301 and

[00119]${.Math.}_{\mathrm{UL}}=2\mathrm{Gbps}$

1301, both across the unlicensed mid-band frequency of f.sub.c.sub.2=5.8 GHz. We also select

[00120]$f_{s_2}{.Math.}_{\mathrm{DL}}=f_{s_2}{.Math.}_{\mathrm{UL}}=2\mathrm{GHz}$

and IF sampling with

[00121]${ℬ}_{{\mathrm{TL}}_2}\stackrel{\Delta }{=}\left[f_{c_2}-f_{s_2}/4,f_{c_2}+f_{s_2}/4\right]\mathrm{and}$${ℬ}_{{\mathrm{ex}}_2}\stackrel{\Delta }{=}\left\{\left\{{ℬ}_{{\mathrm{NBI}}_2}.Math.B_{\mathrm{RB}}.Math.B_{\mathrm{HU}}\right\}.Math.{ℬ}_{{\mathrm{TL}}_2}\right\}.Math.{\stackrel{\_}{ℬ}}_{{\mathrm{TL}}_2}.$ [0209] 3. MTF.sub.3 uses a licensed band at f.sub.c.sub.3=3.5 GHz with

[00122]${.Math.}_{\mathrm{DL}}=14\mathrm{Gbps}$

1301 and

[00123]${.Math.}_{\mathrm{UL}}=6\mathrm{Gbps}$

1301. We also select

[00124]$f_{s_3}{.Math.}_{\mathrm{DL}}=f_{s_3}{.Math.}_{\mathrm{UL}}=14\mathrm{GHz}$

and RF sampling with

[00125]${ℬ}_{{\mathrm{TL}}_3}\stackrel{\Delta }{=}\left[f_{c_3}-f_{s_3}/4,f_{c_3}+f_{s_3}/4\right]\mathrm{and}$${ℬ}_{{\mathrm{ex}}_3}\stackrel{\Delta }{=}\left\{\left\{{ℬ}_{{\mathrm{NBI}}_3}.Math.{ℬ}_{RB}.Math.{ℬ}_{HU}\right\}.Math.{ℬ}_{T{L}_3}\right\}.Math.{\stackrel{\_}{ℬ}}_{T{L}_3}.$

[0210] FIG. 9 displays the PSD, (f) 908, of and the PSD, (f) 809, of 1209 for MTF.sub.3, with nulls at .sub.ex.sub.3, when 1208 is selected as 1214 while 1209 is selected as defined in (16) pre-distorted according to (19). One advantage of MTF.sub.3 over MTF.sub.1 and over MTF.sub.2 is that it is possible to increase

[00126]${\hspace{0.17em}}_3{.Math.}_{\mathrm{DL}}\mathrm{and}{\hspace{0.17em}}_3{.Math.}_{\mathrm{UL}}$

by a multiple by decreasing

[00127]$T_s\stackrel{\Delta }{=}1/f_{s_3}$

by regardless of the value of .sub.m at the cost of an increase in .sub.k.

[0211] Practical Consideration I: In general, it is possible to increase |.sub.UL and |.sub.DL by a multiple while maintaining the same f.sub.s by selecting any combination of the 3 MTF Techniques 1-3 described in Section 5.3.3. For example, by increasing the number of information bits/DOF from 1 bit to 2 bits/DOF, while increasing from 1 to 4, we can have =8, and consequently,

[00128]${.Math.}_{\mathrm{DL}}={.Math.}_{\mathrm{DL}}=1.6\mathrm{Gbps},{.Math.}_{\mathrm{DL}}={.Math.}_{\mathrm{DL}}=16\mathrm{Gbps}\mathrm{and}{.Math.}_{\mathrm{DL}}={.Math.}_{\mathrm{DL}}=112\mathrm{Gbps}.$

[0212] Practical Consideration II: It is possible to increase K by a multiple while keeping |.sub.UL and |.sub.DL fixed and while maintaining the same f.sub.s, by selecting MTF Technique 3 in Section 5.3.3. In this case, the increase in .sub.k is reasonable as long as ≤8. If >8, the increase in .sub.k can remain reasonable by increasing the delay, . For example, it is possible to double from 8 to 16, by doubling from 1 to 2 samples, which halves |.sub.UL and |.sub.DL. The implication of having K increase from 1 to 16 implies that 16 co-located MTF networks can co-exist across the same overlapping licensed and unlicensed bands, after forcing |.sub.UL and |.sub.DL in to be halved.

[0213] Practical Consideration III: It is possible to increase K by >8 without forcing |.sub.UL and |.sub.DL to be halved by decreasing the ratio, K/N.sub.ex, where

[00129]$N_{\stackrel{\_}{\mathrm{ex}}}\stackrel{\Delta }{=}N\left(1-{\zeta }_{\mathrm{ex}}\right),$

as shown below:

[0214] Increasing N.sub.ex when K is fixed: [0215] a) One way to increase N.sub.ex is by increasing at Tx 104, 204, 304, which does not affect the desired DL capacity per MTF device, .sub.DL 1301. It reduces the capacity per MTF device for the UL portion, without affecting the overall desired network capacity, |.sub.UL 1301. [0216] b) Another way to increase N.sub.ex is by increasing indirectly at Rx 108, 208, 308 by taking advantage of the frequency-selective nature of a multipath-rich communication channel 106, 306, 406, which forces N to be replaced by ≥N , or equivalently, forces N.sub.ex to be replaced by

[00130]${\mathrm{��}}_{\stackrel{\_}{\mathrm{ex}}}\stackrel{\Delta }{=}\mathrm{��}\left(1-{\zeta }_{\mathrm{ex}}\right).$

This is often referred to as multipath diversity. In this case, {right arrow over (g)}.sub.l,k 1203, 1206 in (17) is replaced by {right arrow over (g)}.sub.l,k,Ch in (18) at Rx 108, 308. [0217] c) Another way to increase N.sub.ex indirectly at Rx 108, 208, 308 is by increasing the number, N.sub.r, of receive antennas, which does not affect the channel capacity per MTF device, while it increases the overall network capacity accordingly. This is often referred to as spatial diversity or MU-MIMO as shown in FIG. 4. Assuming that each MTF BS/AP contains an array of N.sub.r receiving antennas and that K is the number of simultaneously active MTF devices, therefore, the i.sup.th receiving antenna produces the received signal,

[00131]${\stackrel{.fwdarw.}{r}}_i\stackrel{\Delta }{=}{\stackrel{.fwdarw.}{y}}_i+{\stackrel{.fwdarw.}{w}}_i\forall i\in \left\{1,\mathrm{.Math.},N_r\right\}$

where

[00132]${\stackrel{.fwdarw.}{y}}_i\stackrel{\Delta }{=}{\stackrel{.fwdarw.}{h}}_{i,1,\mathrm{Ch}}*{\stackrel{.fwdarw.}{\beta }}_1+\mathrm{.Math.}{\stackrel{.fwdarw.}{h}}_{i,K,\mathrm{Ch}}*{\stackrel{.fwdarw.}{\beta }}_K;{\stackrel{.fwdarw.}{h}}_{i,j,\mathrm{Ch}}$

is the communications channel 106, 306, 406 between the j.sup.th transmitting antenna and the i.sup.th receiving antenna and {right arrow over (w)}.sub.i is the noise at the i.sup.th receiving antenna.