OVERLAP-SAVE FBMC RECEIVER

Abstract

An FBMC equalization and demodulation unit and corresponding method to process an FBMC signal includes FBMC symbols, each FBMC symbol comprising data mapped over M subcarriers, oversampled by a factor K, filtered by a prototype filter and transposed in the time-domain, comprising: a frequency domain transposition unit, configured to transpose a block of P*KM samples comprising at least one FBMC symbol into frequency domain samples, where P is an integer greater than one, an equalizer unit configured to multiply the frequency domain samples by one or more coefficients computed from a propagation channel estimate, at least one circular convolution unit, configured to perform P circular convolutions between subsets of the equalized samples and a frequency domain response of a frequency shifted version of the prototype filter, and adders, to sum corresponding outputs of each of the P circular convolutions.

Claims

1. An FBMC equalization and demodulation unit, to process an FBMC signal comprising FBMC symbols, each FBMC symbol comprising data mapped over M subcarriers, oversampled by a factor K, filtered by a prototype filter and transposed in the time-domain, the FBMC equalization and demodulation unit comprising: a frequency domain transposition unit configured to transpose a block of P*KM samples comprising at least one FBMC symbol into frequency domain samples, where P is an integer greater than one, an equalizer unit configured to multiply said frequency domain samples by one or more coefficients computed from a propagation channel estimate, at least one circular convolution unit, configured to perform P circular convolutions between subsets of said equalized samples and a frequency domain response of a frequency shifted version of the prototype filter, and adders, to sum corresponding outputs of each of the P circular convolutions.

2. The FBMC equalization and demodulation unit of claim 1, further comprising a down-sampling unit configured to down-sample by a factor K the outputs of the adders.

3. The FBMC equalization and demodulation unit of claim 1, wherein P is chosen so as the signal processed by the frequency domain transposition unit comprises N.sub.s FBMC symbols, with N.sub.s greater or equal to two, the unit further comprising linear phase rotators configured to perform linear phase rotations over the frequency domain samples prior to their processing by the equalizer unit.

4. The FBMC equalization and demodulation unit of claim 3, wherein the linear phase rotation applied over the frequency domain samples is equal to $e^{-j.Math.\frac{.Math.k.Math.n_s}{P}},$ with k an index of the frequency domain sample, n.sub.s[1, N.sub.s] an index of an FBMC symbol in the signal processed by the frequency domain transposition unit.

5. The FBMC equalization and demodulation unit of claim 1, wherein the circular convolution units are numbered from 0 to P1, circular convolution unit number l taking as inputs one out of P outputs of the equalizer unit, starting from output l.

6. The FBMC equalization and demodulation unit of claim 1, wherein G.sub.l, the frequency domain response of a frequency shifted version of the prototype filter used in the circular convolution unit number l, is given by formula: $G_l\left(p\right)=\underset{m=0}{\overset{L-1}{.Math.}}.Math.g\left(m\right).Math.e^{j.Math.\frac{2.Math..Math.l.Math.m}{P.Math.K.Math.M}}.Math.e^{-j.Math.\frac{2.Math..Math.m.Math.p}{K.Math.M}},$ with p[0, L1].

7. The FBMC equalization and demodulation unit of claim 1, wherein the frequency domain transposition unit is configured to perform a Fast Fourier Transform.

8. The FBMC acquisition and demodulation unit of claim 1, wherein the coefficients used by the equalizer unit are computed from a propagation channel estimate using a zero-forcing or a minimum mean-square error technique.

9. The FBMC acquisition and demodulation unit of claim 1, wherein the FBMC signal is transmitted by multiple users, each user being associated to an overlapping factor K.sub.u, an oversampling factor P.sub.u and a number of subcarriers T.sub.u with T.sub.uM, the equalizer unit being configured to take as input P.sub.u*K.sub.uT.sub.u samples depending on the user considered.

10. A receiver comprising the FBMC acquisition and demodulation unit of claim 1.

11. A method for equalizing and demodulating an FBMC signal, the FBMC signal comprising FBMC symbols, each FBMC symbol comprising data mapped over M subcarriers, oversampled by a factor K, filtered by a prototype filter and transposed in the time-domain, the method comprising the steps of: transposing a block of P*KM samples comprising at least one FBMC symbol into frequency domain samples, where P is an integer greater than one, equalizing said frequency domain samples, by multiplying them by one or more coefficients computed from a propagation channel estimate, performing P circular convolutions between subsets of said equalized samples and a frequency domain response of a frequency shifted version of the prototype filter, and summing corresponding outputs of each of the P circular convolutions.

12. A computer program adapted to implement the method of claim 11.

13. A computer readable medium incorporating the computer program of claim 12.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The invention will be better understood and its various features and advantages will emerge from the following description of a number of exemplary embodiments and its appended figures in which:

[0072] FIG. 1 represents the steps of oversampling, filtering and overlapping the symbols in an FBMC transmission of the prior art;

[0073] FIGS. 2a and 2b respectively represent the symbols ordering, and steps of oversampling, filtering and overlapping the symbols in an FBMC/OQAM transmission of the prior art;

[0074] FIGS. 3a and 3b respectively represent a PPN-FBMC/OQAM transmitter and receiver implementation according to the prior art;

[0075] FIG. 4a represents an FS FBMC/OQAM transmitter implementation according to the prior art;

[0076] FIG. 4b describes in details the implementation of an FS FBMC receiver according to the prior art, focusing on the part relative to the equalization and FBMC demodulation;

[0077] FIG. 5a represents an FBMC receiver implementation, where the equalization process is performed prior and independently to the filtering process, as known from the prior art, represented in more detail in FIG. 5b;

[0078] FIGS. 6a and 6b respectively represent an IFFT, and a Decimation In Time decomposition of said IFFT, as known from the prior art;

[0079] FIGS. 7a and 7b represent two embodiments of an FBMC equalization and demodulation unit according to the invention;

[0080] FIG. 8 represents another embodiment of an FBMC equalization and demodulation unit according to the invention, to simultaneously process an entire frame;

[0081] FIG. 9 represents another embodiment of an FBMC equalization and demodulation unit according to the invention, to simultaneously process multiple users; and

[0082] FIG. 10 represents a flow chart of a method FBMC equalization and demodulation method according to an embodiment of the invention.

[0083] The examples disclosed in this specification are only illustrative of some embodiments of the invention.

DETAILED DESCRIPTION OF THE INVENTION

[0084] The invention proposes to modify the processing of the overlap-save FBMC receiver of FIG. 5b, which can hardly be implemented in practice, in order to decrease its complexity. Classical implementations of an overlap-save receiver would consider a P ratio which value is in the order of 1.1 to 4 (depending on the delay spread of the propagation channel), the invention proposes to use a P ratio which value is integer. Indeed, while the complexity of the initial FFT shall suffer from this oversized P ratio, it makes possible to operate major simplifications of the successive computations of the receiver.

[0085] Indeed, by using a P ratio which is an integer, IFFT 530, of size N=PL, can be decomposed using a Decimation In Time method (DIT) into P IFFT of size L plus one additional stage of applying twiddle factors and summing the outputs of the P IFFTs.

[0086] FIG. 6a represents a standard IFFT, taking as inputs N=PL frequency domain samples X(0) to X(N1), transposing these samples to time domain samples x(0) to x(N1). Output x(m) of the IFFT 601 can be expressed as:

[00011]$\begin{array}{cc}x\left(m\right)=\underset{k=0}{\overset{N-1}{.Math.}}.Math.X\left(k\right).Math.e^{j.Math.\frac{2.Math..Math..Math..Math.k.Math.m}{N}},& \left(11\right)\end{array}$

k and m respectively being the indexes of the input and output of the transform.

[0087] FIG. 6b represents the IFFT 601 of FIG. 6a, considering now a Decimation In Time decomposition. The IFFT 601, of a size PL, can be decomposed into P IFFTs 610, 611, of a size L=N/P. The IFFTs are numbered from 0 to P1, IFFT number l takes as inputs the samples X(l) to X(l+(L1).Math.P1) with a step P, and l[0; P1]. Twiddle factors W.sub.l(m) 620, that depends on the index of the calculated output x(m), are applied to the appropriate outputs of the IFFTs, and the results are summed. Output x(m) can be expressed as:

[00012]$\begin{array}{cc}x\left(m\right)=\underset{l=0}{\overset{P-1}{.Math.}}.Math.\left(\underset{k=0}{\overset{L-1}{.Math.}}.Math..Math.X\left(P.Math.k+l\right).Math.e^{j.Math.\frac{2.Math..Math.P.Math.k.Math.m}{N}}\right).Math.e^{j.Math.\frac{2.Math..Math..Math.\mathrm{lm}}{N}},\mathrm{and}& \left(12\right)\\ W_l\left(m\right)=e^{j.Math.\frac{2.Math..Math.\mathrm{lm}}{N}}.& \end{array}$

[0088] Applying the Decimation In Time decomposition to FFT 530 of FIG. 5b, samples x(0) to x(L1), that are to be processed by the filtering stage, can be expressed as:

[00013]$\begin{array}{cc}x\left(m\right)=\underset{l=0}{\overset{P-1}{.Math.}}.Math.\left(\underset{k=0}{\overset{L-1}{.Math.}}.Math.X\left(P.Math.k+l\right).Math.e^{j.Math.\frac{2.Math.\left(\mathrm{Pk}+l\right).Math.\left(m+l_0\right)}{N}}\right),.Math..Math.m\left[0,L-1\right].Math..Math.\mathrm{and}& \left(13\right)\\ x.Math.\left(m\right)=\underset{l=0}{\overset{P-1}{.Math.}}.Math.\left(\underset{k=0}{\overset{L-1}{.Math.}}.Math.X\left(P.Math.k+l\right).Math.e^{j.Math.\frac{2.Math.\left(\mathrm{Pk}+l\right).Math.l_0}{N}}.Math.e^{j.Math.\frac{2.Math..Math.k.Math.m}{L}}\right).Math.e^{j.Math.\frac{2.Math..Math..Math.l.Math..Math.m}{N}},.Math..Math.m\left[0,L-1\right]& \left(14\right)\end{array}$

[0089] In what follows, the following notations shall be used:

[00014]$\begin{array}{cc}x\left(m,l\right)=\underset{k=0}{\overset{L-1}{.Math.}}.Math.X\left(P.Math.k+l\right).Math.e^{j.Math.\frac{2.Math.\left(\mathrm{Pk}+l\right).Math.l_0}{N}}.Math.e^{j.Math.\frac{2.Math..Math.k.Math.m}{L}}.Math..Math.\mathrm{and}& \left(15\right)\\ X\left(k,l\right)=X\left(P.Math.k+l\right).Math.e^{j.Math.\frac{2.Math.\left(\mathrm{Pk}+l\right).Math.l_0}{N}}& \left(16\right)\end{array}$

[0090] The next stage of the FBMC receiver of FIG. 5b consists in the filtering 540 which corresponds to windowing the samples x(m) by the prototype filter g(m), and the transposition of the output of the filtering to the frequency domain through FFT 550 of a size L.

[0091] Considering the previous equations, the outputs Y(k) of FFT 550 can be expressed as:

[00015]$\begin{array}{cc}Y\left(k\right)=\underset{l=0}{\overset{P-1}{.Math.}}.Math.\underset{m=0}{\overset{L-1}{.Math.}}.Math.g\left(m\right).Math.x\left(m,l\right).Math.e^{j.Math.\frac{2.Math..Math..Math.l.Math..Math.m}{N}}.Math.e^{-j.Math.\frac{2.Math..Math.k.Math.m}{L}},k\left[0,L-1\right]& \left(17\right)\end{array}$

[0092] It can therefore be deduced that each output of FFT 550, following the filtering stage 540, can be seen as the sum of P equivalent FBMC filters, wherein each equivalent FBMC filter processes the outputs of a size L IFFT multiplied by a linear phase rotation term

[00016]$e^{j.Math.\frac{2.Math..Math..Math.l.Math..Math.m}{N}}.$

[0093] The linear phase rotation term

[00017]$e^{j.Math.\frac{2.Math..Math..Math.l.Math..Math.m}{N}}$

can be seen as a carrier-frequency offset which can be integrated in frequency domain thanks to the time and frequency localization of the prototype filter:

[00018]$\begin{array}{cc}Y\left(k\right)=\underset{l=0}{\overset{P-1}{.Math.}}.Math.\left(\underset{p}{.Math.}.Math.G_l\left(p\right).Math.\left(\underset{m=0}{\overset{L-1}{.Math.}}.Math.x\left(m,l\right).Math.e^{-j.Math.\frac{2.Math.\left(k-p\right).Math.m}{N}}\right)\right),.Math.k\left[0,L-1\right],\mathrm{with}.Math.\text{:}& \left(18\right)\\ G_l\left(p\right)=\underset{m=0}{\overset{L-1}{.Math.}}.Math.g\left(m\right).Math.e^{j.Math.\frac{2.Math..Math..Math.\mathrm{lm}}{N}}.Math.e^{-j.Math.\frac{2.Math..Math.m.Math.p}{L}}& \left(19\right)\end{array}$

and the set of coefficient indexes where the frequency response of the frequency shifted prototype filter is not null, the coefficient indexes belonging to the interval [0, L1].

[0094] The filtering stage can be computed using a circular convolution operation with the filter coefficients G.sub.l(p). At this point, IFFT 530 is decomposed in P IFFT of size L, which outputs are directly connected to the inputs of P FFT equivalent to the FFT 550, which also are of size L. Thus, all the FFT and IFFT of size L can be removed from the implementation, which results in the equation:

[00019]$\begin{array}{cc}Y\left(k\right)=\underset{l=0}{\overset{P-1}{.Math.}}.Math.\left(\underset{p}{.Math.}.Math.G_l\left(p\right).Math.X\left(k-p,l\right)\right),k\left[0,L-1\right],& \left(20\right)\end{array}$

[0095] Coefficients G.sub.l may be computed by: [0096] considering the impulse response of the prototype filter, [0097] applying to the impulse response a linear phase rotation of

[00020]$e^{j.Math.\frac{2.Math..Math..Math.l.Math..Math.m}{N}},$

and [0098] applying a Fourier Transform to the rotated impulse response.

[0099] Advantageously, thanks to the frequency localization of the prototype filter, these coefficients may be simplified by setting to zero all the coefficients which value is not significant, i.e. which value is below a certain percentage of the highest coefficient, for instance 1%.

[0100] Therefore, in the FBMC receiver according to the invention, the processing of FIG. 5b of: [0101] performing an IFFT of a size N, [0102] discarding the useless outputs 531 of IFFT 530, [0103] filtering 540 the remaining outputs of IFFT 530 by filter g(m), [0104] transposing the result of the filtering in the frequency domain using FFT 550,
can be replaced by: [0105] P circular convolutions between specific outputs of FFT 510 and vectors G.sub.l, which are frequency domain transforms of the prototype filter g(m) to which carrier-shift rotations are applied, and [0106] sums between the outputs of the circular convolutions.

[0107] FIG. 7a represents an embodiment of an FBMC equalization and demodulation unit according to the invention.

[0108] An initial FFT 710 of a size N=PL, with L the size of an oversampled FBMC symbol and P an integer greater than one, is performed over samples including the FBMC symbol of interest. The aim of the FFT is to convert the received signal into the frequency domain. Using a FFT is advantageous in terms of implementation, but a discrete Fourier Transform would provide the same results. The outputs of FFT 710 are equalized 720, using either a single tap or a multi-taps equalizer. Contrary to FIG. 5b, once equalized, the samples in the frequency domain are not transposed back in the time domain, but are processed by P units (730, 731) performing circular convolutions of a size L.

[0109] An index l, l[0,P1], is attributed to each unit performing circular convolution. Unit number l takes as input the outputs number l, l+P, l+2P, . . . , l+(L1)P, and performs a convolution between said inputs, and vector G.sub.l, where G.sub.l(p) is computed from the prototype filter according to the already described equation:

[00021]$\begin{array}{cc}{G}_l\left(p\right)=\underset{m=0}{\overset{L-1}{.Math.}}.Math.g\left(m\right).Math.e^{j.Math.\frac{2.Math..Math..Math.\mathrm{lm}}{N}}.Math.e^{-j.Math.\frac{2.Math..Math.m.Math.p}{L}}& \left(21\right)\end{array}$

[0110] To calculate sample Y(k), the k.sup.th output of each of the circular convolution units are summed (740). Those Y(k) are then down-sampled by a factor K by a final down-sampler 750.

[0111] FIG. 7b represents another embodiment of an FBMC equalization and demodulation unit according to the invention, in which the final stage of down-sampling has been removed. Indeed, this stage can easy be suppressed by simply calculating the Y(k) (771, 772, 773, 774) for subcarriers that are to be kept after the down-sampling stage, namely subcarriers k that are a multiple of the overlap factor K.

[0112] The FBMC equalization and demodulation unit according to the invention is compatible with any FBMC scheme (QAM, OQAM or any other), and can be implemented for any prototype filter length. As based over an overlap-save technique, it responds to the lack of cyclo-stationarity related to the use of short prototype filters, such filters being mandatory to achieve low latency transmissions.

[0113] In the prior art implementations, like for instance in the FS-FBMC receiver of FIG. 4b or the overlap-save FBMC receiver of FIG. 5b, FFTs 460 and 550 are specifically of a size L=KM. The implementation is therefore limited to the use of a single type of prototype filter, and multiple FBMC processing chains must be implemented in the receiver chain to handle various prototype filters, each prototype filter being defined by a size L, an overlapping factor K and coefficients G.sub.l. In addition, since L=KM, multiple subcarrier spacing (related to M) cannot be supported with prior art implementations. In the FBMC equalization and demodulation unit according to the invention, multiple prototype filters can be supported as long as N is a multiple of L for each prototype filter. Indeed, the architecture of the convolutions performing the filtering only depend on the prototype filter having the highest number of coefficients G.sub.l. For prototype filters having a lower number of coefficients, null coefficients are added to comply with this architecture. Furthermore, the number of circular convolution units (P) to perform only depends on the prototype filter size L, since N is fixed, and N=PL. Thus, multiple filter size can be supported by enabling or disabling the processing of circular convolution units, depending on the number of required convolutions to perform the demodulation. In addition, multiple overlapping factors K can be supported by simply adapting the downsampling unit 750. Thus, a same receiver implementation can process multiple FBMC schemes (size of prototype filter, overlapping factor and/or coefficients) as long as N=PL with P an integer. The FBMC receiver according to the invention therefore fits the requirement of adaptability expected from 5G communications.

[0114] The FBMC equalization and demodulation unit according to the invention can be implemented over multiple hardware/software architectures.

[0115] Among the various possible implementations, the various units required (FFT, circular convolution units, summers) can be embedded over a single calculation machine such as a software reprogrammable calculation machine (microprocessor, microcontroller, digital signal processor (DSP), graphics processing unit (GPU), . . . ), a dedicated calculation machine (Field Programmable Gate Array (FPGA), Application Specific Integrated Circuit (ASIC), . . . ), or any other appropriate equipment.

[0116] They can also be implemented by means of computer-application programs or services, as an application-programming interface (API), a library, and/or other computer-program product, or any combination of such entities.

[0117] The subject matter of the present disclosure includes all novel and non-obvious combinations and sub-combinations of the various processes, systems and configurations, and other features, functions, acts, and/or properties disclosed herein, as well as any and all equivalents thereof.

[0118] The FBMC equalization and demodulation unit according to the invention may be embedded in a receiver, receiving the signal from an antenna and a RF chain in charge of converting the signal to an intermediate frequency or to baseband, and delivering equalized and demodulated data to a unit in charge of computing the subsequent algorithms required to receive the data transmitted, as for instance a QAM or OQAM demodulator, an error code decoder, and/or the functions of the OSI layers that are above the PHY layer. It may also be embedded in a standalone device configured to take as input an intermediate frequency or baseband signal, and to provide equalized and demodulated data to another reception device.

[0119] Compared to the implementation of FIG. 5b, the FBMC equalization and demodulation unit implementation according to the invention saves one IFFT 530 of a size N=PL, one FFT 550 of a size L, and one filter 540 of a length L, at the expense of P circular convolution units 730, 731, and L adders 740.

[0120] The receiver according to the invention can be implemented by parallelizing the circular convolutions, to optimize the performances, or using in sequence a single circular convolution unit, to optimize the implementation cost. In addition, when the scheme considered is an FBMC/OQAM scheme, only the real (or imaginary) part of the signal is processed at the output of the circular convolutions. Since the filter's coefficients are constants, the multiplications can be implemented by way of adders only, which further reduces the implementation complexity.

[0121] Vectors G.sub.l, used by the circular convolutions, do not have to be calculated for each iteration, and can be a set of parameters associated to a specific FBMC scheme stored in a memory that can be accessed by the circular convolution units.

[0122] FIG. 8 represents another embodiment of an FBMC equalization and demodulation unit according to the invention, in which the initial FFT 810 is performed over a block comprising multiple FBMC symbols. In the example of FIG. 8, the symbols are transmitted using an FBMC/OQAM scheme, so that successive FBMC symbols are transmitted over either the real or imaginary part of the QAM samples, and shifted from half an FBMC symbol. In this illustrative example, P is chosen as being higher than or equal to four, so that the block of samples processed by the initial FFT comprises seven FBMC/OQAM symbols, seven symbols being a good compromise between complexity, latency and spectral efficiency. This embodiment may be easily transposed to various FBMC schemes, using QAM or any other data mapping over the subcarriers, other number of symbols considered by the initial FFT, or other block sizes.

[0123] In the embodiment of FIG. 8, the equalized and demodulated samples Y(k) are calculated for each of the FBMC symbols comprised in the block considered by FFT 810 of a N=PL size. For this purpose, a linear phase rotation 811 is applied to the frequency domain samples between FFT 810 and the equalization stage 820. This linear phase rotation is a circular time shift applied to the receiver frequency domain samples, and is calculated depending on the symbol considered. The phase rotation to apply to each subcarrier is

[00022]$W\left(k\right)=e^{-j.Math.\frac{2.Math..Math.k.Math.n_s.Math.M}{2.Math.N}},$

which is equal to

[00023]$e^{-j.Math.\frac{.Math.k.Math.n_s}{P.Math.K}},$

where n.sub.s[1, N.sub.s] is the index of the FBMC symbol in the block of data processed, N.sub.s being the number of FBMC symbols in the block of data processed.

[0124] The subsequent stages of the FBMC equalization and demodulation unit are the same as those of FIG. 7a: [0125] a stage 820 of equalizing the outputs of FFT 810, [0126] a stage of performing P circular convolutions 830, 831 between outputs of FFT 810 and vectors G.sub.l, [0127] a stage of summing 840 outputs of the P circular convolutions to calculate equalized and demodulated samples Y(k), and [0128] a stage 850 of down-sampling Y(k) by a factor K.

[0129] As illustrated in FIG. 7b, the stage 850 of down sampling can be advantageously suppressed by directly calculating the down-sampled samples Y.sub.DS(k) during the stage 840 of summing outputs of the circular convolution units.

[0130] Advantageously, the linear phase shift 811 can be applied along with the equalization in a single stage 812 of phase shifting and equalizing. Therefore, the phase shifts applied to output X(k) in this stage are equal to W(k)(k).

[0131] This embodiment of an FBMC receiver implementation according to the invention only uses one FFT to demodulate all the FBMC symbols in the transmitted block, reducing thus the computational complexity.

[0132] In order to quantify the gain provided by the invention in terms of implementation complexity, some measurements have been done over an FBMC/OQAM scheme with different values of prototype filter lengths M and an overlap factor of K=2. The measurements have been done over a standard FS-FBMC receiver implementing an overlap-save technique as described in FIG. 5a, the embodiment of the invention presented in FIG. 7b, and the embodiment of the invention presented in FIG. 8.

[0133] In the first scenario, M=2048:

TABLE-US-00001 Number of multipliers required Ratio to per FBMC symbol FS-FBMC FS-FBMC (FIG. 5a) ~248000 1 Embodiment of FIG. 7b ~97500 0.39 Embodiment of FIG. 8 ~27300 0.11

[0134] In the second scenario, M=512:

TABLE-US-00002 Number of multipliers required Ratio to per FBMC symbol FS-FBMC FS-FBMC (FIG. 5a) ~47500 1 Embodiment of FIG. 7b ~20300 0.74 Embodiment of FIG. 8 ~6241 0.13

[0135] It can be seen from the above measurements that, depending on the embodiment of the invention considered, the implementation cost of the FBMC receiver can be reduced of almost 90% compared to the prior art.

[0136] It must also be noticed that, as the FBMC receiver implementation according to the invention complies with various FBMC symbol sizes, it is well suited for multi-user transmissions, where different subcarriers are allocated to different users potentially using different FBMC schemes (number of subcarriers allocated and overlap factor).

[0137] In what follows, it is considered that, for each user u, a group of T.sub.u subcarriers is allocated among M.sub.u representing the total number of subcarriers (allocated or not). Generally, M.sub.u is related to the subcarrier spacing F.sub.u=M.sub.u/F.sub.s, where F.sub.s is the sampling frequency, assumed to be the same for all users. Furthermore, an overlap factor K.sub.u is considered for each user u. As a result, they use different prototype filters, each prototype filter having a length equal to L.sub.u=M.sub.uK.sub.u samples, and having a frequency response G.sub.u which corresponds to .sub.u coefficients. The number of coefficients .sub.u may be advantageously reduced by considering only the significant coefficients of G.sub.u. At the receiver side, there is no interference between users if they transmit data on separate frequencies, thanks to the frequency localization of the prototype filter.

[0138] With an FBMC receiver according to the prior art, processing multiple users using different FBMC schemes requires the implementation of one receiver per scheme. Considering for instance the PPN-FBMC receiver of FIG. 3b, the filtering is realized in the time domain, before the FFT. Thus, one implementation per prototype filter is required to process FBMC signals having different schemes. Considering the FS-FBMC receiver of FIG. 4b, the size of the FBMC symbols considered has a direct impact over the size of FFT 460, which must be duplicated for each FBMC scheme. Considering the overlap-save receiver of FIG. 5b, the equalizer part may be common to various FBMC schemes, but the FBMC demodulator part 508 depends on the size of the FBMC symbols.

[0139] The invention allows processing each user considering a single receiver implementation, which drastically reduces its implementation cost.

[0140] FIG. 9 represents another embodiment of an FBMC equalization and demodulation unit implementation according to one embodiment of the invention, prone to process signals coming from different users using (or not) various FBMC schemes.

[0141] This receiver comprises a first FFT 910, which size N is an integer multiple of all the prototype filters lengths: u,mod.sub.L.sub.u(N)=0, which means that, whatever u, there is a P.sub.u that satisfies the formula N=P.sub.uL.sub.u.

[0142] The FBMC multi-user receiver according to the invention processes each user independently (920, 930), but may use the same FBMC receiver implementation for each FBMC scheme. To this end, the FBMC receiver implementation has to be configured depending on the parameters of each user transmission as follows: [0143] P.sub.u circular convolutions are to be considered. As seen previously, depending on the embodiment, the P.sub.u circular convolutions can be processed by one circular convolution unit performing in sequence the P.sub.u circular convolutions, by P.sub.u circular convolution units working in parallel, or by a combination thereof; [0144] the P.sub.u circular convolutions are configured to process the sets of coefficients G.sub.l,u, with l an index of the circular convolution. The G.sub.l,u are computed from the prototype filter, or retrieved from a memory, and are the frequency domain response of a frequency shifted version of the prototype filter; [0145] the down-sampling factor must be set to K.sub.u.

[0146] These modifications are different parameters of a same implementation, which comprises, in addition to the FFT 910, a stage of selecting a certain number of subcarriers attributed to the user considered, and the stages of performing a frequency domain equalization over said subcarriers (921), performing P.sub.u circular convolutions over the equalized samples (922, 923), adding the corresponding outputs of the P.sub.u circular convolutions (924), and down-sampling the result by a factor K.sub.u (925). The down-sampling may be avoided by adding only one over K.sub.u outputs of the P.sub.u circular convolutions. The step of selecting a certain number of subcarriers attributed to the user considered is done by selecting the P.sub.uK.sub.uT.sub.u upsampled subcarriers that correspond to the original T.sub.u subcarriers allocated to the user concerned.

[0147] The FBMC multi-user receiver according to the invention can be used to demodulate FBMC symbols for all users sharing the lowest total number of subcarriers M.sub.min=min(M.sub.u), referred to as the elementary FBMC symbol. Then, the FBMC symbols having a higher value of M.sub.u can be demodulated each M.sub.u/M.sub.min elementary FBMC symbols. It is assumed that M.sub.u/M.sub.min are integers, which is generally the case because otherwise, it complicates the frame structure.

[0148] Advantageously, when the frame length is the same for each user, the FBMC receiver according to the invention can be combined with the embodiment presented in FIG. 8, in order to process all the FBMC symbols of a frame for each user considering only one FFT 910.

[0149] The invention further addresses a corresponding method, to equalize and demodulate an FBMC signal in a receiver. FIG. 10 represents a flow chart of such a method according to an embodiment of the invention.

[0150] The method is to be processed over an FBMC signal, where samples are mapped over M subcarriers. Among the subcarriers, some are dedicated to the mapping of the data symbols, while some others are dedicated to the mapping of pilot sequences, or are left empty (guard subcarriers or unused subcarriers). The M subcarriers are transposed in the time domain, oversampled by a factor K, and filtered by a prototype filter. The method according to the invention shows good performances whatever the size of the prototype filter and the oversampling ratio, and is compliant with various FBMC symbol sizes as long as the size of the initial FFT, performed during the first step of the method, is a multiple of the prototype filter length. Preliminary steps of synchronizing the receiver in time and frequency over the received signal and calculating a propagation channel estimate and the corresponding vector required to equalize the signal, which are not part of the method object of the invention, are required and considered as realized.

[0151] The method comprises a first step 1001 of transposing in the frequency domain a block of received signal, said block comprising the FBMC symbol which equalization and demodulation is to be performed. This transposition shall advantageously be realized considering a Fast Fourier transform, for implementation purposes. The size of the block is of N=PKM, where P is an integer greater than one. The method further comprises a step 1002 of equalizing the frequency domain samples by multiplying each frequency domain sample by one or more coefficients computed from a propagation channel estimate. Advantageously, the coefficients may be computed using a zero-forcing or a minimum mean-square error technique.

[0152] The method further comprises a step 1003 of performing P circular convolutions between subsets of equalized samples, and a frequency domain response of a frequency shifted version of the prototype filter. Each unit performing circular convolution uses a distinct set of inputs. For instance, circular convolution unit number l takes as input one output of the equalizer unit out of P, starting from output l. Each unit performing circular convolution between a set of equalized samples and a distinct set of coefficients. For instance, circular convolution unit number l correlates the equalized samples with the frequency domain response of a frequency shifted version of the prototype filter G.sub.l given by formula:

[00024]$\begin{array}{cc}{G}_l\left(p\right)=\underset{m=0}{\overset{L-1}{.Math.}}.Math.g\left(m\right).Math.e^{j.Math.\frac{2.Math..Math..Math.\mathrm{lm}}{N}}.Math.e^{-j.Math.\frac{2.Math..Math.m.Math.p}{L}},& \left(22\right)\end{array}$

with p[0, L1].

[0153] The method also comprises a step 1004 of summing outputs of the P circular convolution units, to compute the equalized and demodulated samples Y (k). Y(k) is obtained by summing the outputs k of the P circular convolutions units. The Y(k) are then down-sampled by a factor K to retrieve down-sampled samples Y.sub.DS(k) Advantageously, the samples Y.sub.DS(k) may be retrieved by calculating the samples Y(k) only for indexes k that are multiple of K.

[0154] In another embodiment of the FBMC equalization and demodulation process according to the invention, an additional step may be added. In this embodiment, the block of samples processed by the frequency transposition unit is selected so as to comprise multiple FBMC symbols. In this embodiment, a single frequency transposition is required to equalize and demodulate all the FBMC symbols comprised in the block of samples processed, which is advantageous in terms of processing power required.

[0155] In this embodiment, a step 1005 of applying a linear transposition to the equalized samples is added, previous to the circular convolution processing. The linear transposition applied to equalized sample k is

[00025]$e^{-j.Math.\frac{.Math.k.Math.n_s}{P.Math.K}},$

with n.sub.s [1, N.sub.s] the index of the FBMC symbol considered in the block of signal processed, and N.sub.s the number of FBMC symbols comprised in the block of signal processed.

[0156] The method according to the invention may be used to process multiple users, by considering, for each user, a number of circular convolutions and the parameters of the circular convolutions that depend on the prototype filter.

[0157] The method according to the invention may take the form of a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or an instruction execution system. A computer-usable or computer-readable can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The medium can be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation medium.

[0158] While embodiments of the invention have been illustrated by a description of various examples, and while these embodiments have been described in considerable details, it is not the intent of the applicant to restrict or in any way limit the scope of the appended claims to such details. The invention in its broader aspects is therefore not limited to the specific details, representative methods, and illustrative examples shown and described.