Mode division multiplexed passive optical network

Abstract

A method of compensating for crosstalk in a mode division multiplexing passive optical network using a technique of transmitter-side crosstalk pre-compensation, performed at the Central Office, in which a downlink reference signal such as a training sequence or pilot signal is retrieved at the transmitter without being influenced by crosstalk effects on its uplink transmission. An uplink reference signal is transmitted in a quasi-single mode transmission along the optical fiber, and a plurality of optical signals input to transmission multiplexer are adapted based on the uplink reference signal to pre-compensate for crosstalk.

Claims

1. A mode division multiplexing passive optical network comprising: a plurality of input channels, each of the plurality of input channels being arranged to convey an input optical signal in a different one of a plurality of modes; an optical transfer unit comprising: a multiplexer having a plurality of input ports and an output port, wherein each of the plurality of input ports is connected to a respective one of the plurality of input channels; an optical fibre having an uplink end connected to the output port, wherein the optical fibre is arranged to receive from the multiplexer a mode multiplexed signal corresponding to the input optical signals from the plurality of input channels; and a demultiplexer having an input port and a plurality of output ports, wherein the input port is connected to a downlink end of the optical fibre, and wherein the demultiplexer is arranged to divide the mode multiplexed signal between each of the plurality of output ports; and a plurality of output channels, each of the plurality of output channels being connected to a respective one of the plurality of output ports and being arranged to convey an output optical signal in a different one of the plurality of modes conveyed by the plurality of input channels, wherein the optical transfer unit is arranged to: transmit a plurality of downlink reference signals in the mode multiplexed signal, each of the plurality of downlink reference signals being in a different one of the plurality of modes, receive the plurality of downlink reference signals at the plurality of output ports, couple the received plurality of downlink reference signals into an uplink reference signal, transmit the uplink reference signal in a quasi-single mode transmission along the optical fibre, and receive the uplink reference signal at the uplink end of the optical fibre; and wherein the passive optical network further comprises a downlink signal pre-compensation module arranged to adapt the optical signals on two or more of the plurality of input channels to pre-compensate for crosstalk in the optical transfer unit, wherein the downlink signal pre-compensation module is controllable based on the received uplink reference signal.

2. The network of claim 1, wherein the multiplexer is mode-selective.

3. The network of claim 2, wherein the downlink reference signals are pilot tones transmitted on degenerate mode pairs in the plurality of modes.

4. The network of claim 2, wherein the downlink pre-compensation module comprises an electrical butterfly FIR filter on each of the two or more of the plurality of input channels.

5. The network of claim 1, wherein each downlink reference signal comprises a training sequence transmitted in a downlink training sequence time slot.

6. The network of claim 5, wherein the training sequence time slot includes a guard band having a duration long enough to encompass a uplink training sequence that comprises the received downlink reference signals in series.

7. The network of claim 1, wherein the uplink reference signal is transmitted using a non-degenerate mode in the plurality of modes.

8. The network of claim 1, wherein the optical fibre is a few mode fibre (FMF) capable of supporting propagation in a plurality of degenerate and non-generate modes.

9. A method of compensating for crosstalk in a mode division multiplexing passive optical network, the method comprising: inputting a plurality of downlink reference signals to a multiplexer, each of the plurality of downlink reference signals being in a different one of a plurality of modes; transmitting the plurality of downlink reference signals as a mode multiplexed signal along an optical fibre; receiving the mode multiplexed signal at a demultiplexer; outputting a plurality of received downlink reference signals from the demultiplexer, each of the received plurality of downlink reference signals being in a different one of a plurality of modes; coupling the received plurality of downlink reference signals into an uplink reference signal, transmitting the uplink reference signal in a quasi-single mode transmission along the optical fibre, adapting a plurality of optical signals input to the multiplexer based on the uplink reference signal to pre-compensate for crosstalk.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Examples of the inventions are discussed below with reference to the accompanying drawings, in which:

(2) FIG. 1 is a schematic diagram of a known architecture for a mode division multiplexing passive optical network (MDM-PON), which is discussed above;

(3) FIG. 2 is an illustration of section I of a fibre model transfer matrix, also discussed above;

(4) FIG. 3 is a schematic diagram of an MDM-PON downlink architecture that is an embodiment of the present invention;

(5) FIG. 4 is a graph showing error free bandwidth as function of crosstalk;

(6) FIG. 5 is a schematic diagram of an MDM-PON downlink architecture that is another embodiment of the present invention;

(7) FIG. 6 is a schematic diagram of a downlink PON frame that is an idealized representation of the training sequences at different points of the proposed architecture;

(8) FIG. 7 is a graph illustrating performance of the MDM-PON downlink architecture shown in FIG. 5; and

(9) FIG. 8 is a schematic diagram showing configurations for a multiplexer and a demultiplexer suitable for use with the invention.

DETAILED DESCRIPTION

(10) The present invention is based on the recognition that, in typical FMFs, the differential mode delay between degenerate modes is usually very low (?1 ps/km). This means that matrix terms of XT.sub.i relating these modes (e.g. relating LP11a and LP11b or LP21a and LP21b) have a low dependency on the frequency. These terms are primarily responsible for the introduction of crosstalk as explained above.

(11) The crosstalk characteristics of the mode multiplexer 108 and mode demultiplexer 112 shown in FIG. 1 can be described by a respective unitary matrices (e.g. H.sub.MUX and H.sub.DEMUX). After inversion, these matrices can be used to fully compensate for the mode mixing. However, by appropriate design of the mode multiplexer 108 and mode demultiplexer 112 known to a skilled person, the crosstalk introduced between non-degenerate LP modes can be reduced to less than ?20 dB, which means it can be treated as negligible for the purposes of the invention. In this case, only the crosstalk between degenerate LP modes needs to be compensated.

(12) In this invention, we propose the mitigation of the crosstalk introduced along the FMF either by using mode selective MUX/DEMUX and mitigating only the crosstalk taking place between degenerate modes, i.e. the terms highlighted in boxes 201, 202, 203, 204 in FIG. 2, or by using mode non-selective MUX/DEMUX and mitigating the crosstalk between the full set of modes.

(13) FIG. 3 shows the proposed downlink architecture using mode selective MUX/DEMUXs. In this case, the crosstalk can be mitigated by compensating only for the crosstalk taking place between the degenerate modes. Furthermore, it can be shown that because the DMD between degenerate modes is usually very low (?1 ps/km), the crosstalk terms between degenerate modes show low dependency on the frequency. Thereby, the crosstalk terms can be estimated by using pilot tones (in-band and/or out-of-band).

(14) The electrical channel estimation/inversion at the central office for the downlink is based on pilot tones (PT) that are retrieved after the mode selective DEMUX at the ODN before the optical splitters, as shown in FIG. 3. However, the signals at the mode DEMUX cannot be simply mirrored back since they would undergo through additional crosstalk along the uplink. To overcome this issue, the invention proposes the usage of quasi-single mode transmission for the retrieval of the pilot tones using one (or more) of the non-degenerate modes (LP01, LP02, LP0x, . . . ). In other words, the outputs from the mode DEMUX which correspond to pairs of degenerate modes are tapped (e.g. using a 80:20 beam splitter), coupled together, and transmitted back to the CO over one (or more) of the non-degenerate modes (LP01, LP02, . . . ). Note that the usage of more than one of the LP0x modes allows to reduce the coupling losses (see point M in FIG. 3). For instance, for 6 modes (LP01, LP02, LP11a, LP11b, LP21a, LP21b), the pilot tones of the pair LP11a/LP11b can be transmitted over LP01 and the pilot tones of the pair LP21a/LP21b can be transmitted over LP02. In this way, two 2?1 couplers can be used instead of one 4?1 coupler (the losses of the coupler scale with the number of ports). For more than 6 modes, similar reasoning applies.

(15) The channel estimation can be carried out relying on one pilot tone per pair of degenerate modes or two pilot tones with different frequencies per pair of degenerate modes. Moreover, in order to avoid the interference between pilot tones when they are coupled together after the mode DEMUX, unique frequencies can be addressed to each pair of degenerated modes.

(16) Once the transmitted PTs are directly detected at the central office, these are used for the blind estimation of the 2?2 matrices that describe the mode coupling LP11a and LP11band the mode coupling between LP21a and LP21b. The subsequent information symbols are pre-equalized with this estimated matrices in such a way the information can be successfully recovered at the ONUs.

(17) In order to test the configuration proposed in FIG. 3, the performance of a MDM system supporting 3 modes (LP01, LP11a, LP11b) and one 10 Gbps intensity modulated direct detected (IM/DD) system per mode, operating independently from each other, was modelled and simulated. The crosstalk introduced by the fibre was varied from ?40 dB/km to ?15 dB/km, and the mode MUX/DEMUX was assumed to introduce ?20 dB crosstalk. The fibre was assumed to introduce a differential mode delay of 1000 ps/km between LP01 and LP11a/b and 1 ps/km between LP11a and LP11b. Finally, in order to assess the frequency dependency of the fibre transfer matrix coefficients, the estimation was done at 1547.5 nm and the 10 Gbps IM/DD channels where transmitted at a different wavelength that was varied from 1530 nm to 1565 nm.

(18) FIG. 4 shows the error free bandwidth as function of the FMF crosstalk. The results show that for a fibre presenting a crosstalk strength of ?25 dB/km, the method proposed would be able to compensate the crosstalk between the degenerate modes over 5 nm. On the other hand, for a fibre with a crosstalk around ?34 dB/km, the method proposed would be able to compensate the crosstalk between the degenerate modes over 32 nm (the whole extended C-band). Table 1 present a list of properties of several FMFs presented in the literature. It can be see that they present a crosstalk strength of between ?30 and ?40 dB/km. The present invention can thus be expected to enable meaningful compensation using such FMFs.

(19) TABLE-US-00001 TABLE 1 Properties of known FMFs XT LP Refractive- Fibre [dB/km] modes Index Profile Reference [1] ?39.77 3 Graded-Index L. Gr?ner-Nielsen, et al., J. Lightw. Technol., 30(23), p. 3693, 2012. [2] ?36.53 3 Step-Index A. Li, et al., Proc. OFC2011, p. PDPB8. [3] ?33.19 3 Graded-Index R. Ryf, et al., J. Lightw. Technol., 30(4), p. 521, 2012. [4] ?30.21 6 Graded-Index T. Mori, et al., Proc. OFC2013, p. OTh3K.1.

(20) FIG. 5 shows a proposed downlink architecture using mode non-selective MUX/DEMUXs. In this case, the crosstalk can only be mitigated by compensating for the full set of modes. Furthermore, because the DMD between non-degenerate modes is usually high (10-100 ps/km and higher), the channel matrix is frequency dependent. Thereby, pilot tones cannot be used. Instead it is proposed to use training sequences to perform channel estimation. Note that the architecture in FIG. 5 would be equally valid when using mode selective MUX/DEMUX. However, in that case, the usage of pilot tones is possible and the architecture in FIG. 5 could be simplified to match the architecture presented in FIG. 3, as explained above.

(21) The electrical channel estimation/inversion at the central office for the downlink is based on training sequences that are retrieved after the mode non-selective DEMUX at the ODN before the optical splitters, as shown in FIG. 5.

(22) After downstream propagation, the downlink signals are tapped just after the mode DEMUX (see FIG. 5) so that they can be returned to the central office and be used to estimate the crosstalk characteristic of the optical transfer unit. Similarly to FIG. 3, it is proposed to use quasi-single mode transmission for the retrieval of the training sequences (using non-degenerate modes LP01, LP02,). However, unlike the downlink reference signals, which can be sent in parallel, the uplink reference signal must send the training sequences in series if a quasi-single mode transmission is used. Thus, the tapped signals are added together to form the uplink reference signal after going through different delays. In order to be able to accommodate such delays, time guard bands are included around the training sequences. FIG. 6 shows an idealized representation of the training sequences at different points of the architecture shown in FIG. 5. Once the transmitted TS are detected in the OLT, these are processed in order to get an estimation of the multiple-input multiple-output (MIMO) matrix (a 6?6 matrix in this embodiment) which describes the downlink channel, including the mode coupling and delay from the few-mode fibre. The subsequent information symbols are pre-equalized with this estimated matrix in such a way the information can be successfully recovered at the ONUs.

(23) When the signals propagate through the few-mode fibre, they undergo mode coupling and dispersion mode delay, whose effects must be effectively captured by the training sequences. A number of different suitable types of training sequences can be used to get an estimation of the MIMO channel matrix. In one embodiment, constant-amplitude zero-autocorrelation (CAZAC) sequences can be used. Taking advantage of their shift-orthogonality property, the length of each training sequence sample must fulfil the following:
T.sub.TS?N.Math.?T.sub.ch+G1

(24) where ?T.sub.ch is the time broadening due to dispersion mode delay the signals undergo in the downlink direction, and G is additional time duration to accommodate corrupted signal samples due to the dispersion mode delay-induced broadening. In this way, the first term of the right hand side member of equation 1 (N.Math.?T.sub.ch) aims to get a sufficient number of signal samples for the estimation of the MIMO downlink channel, whilst the second term (G) aims to protect these samples from getting corrupted when the signals propagate in the downlink and uplink directions. Thus, the value of this additional time duration must be set taking into account that the signals have propagated through the few-mode fiber twice (one when going from point A to point B, and one more when going from point C to point D in FIG. 5). Therefore, G is given by:
G=2?T.sub.ch2

(25) Although a smaller value of G might be possible, it could lead to a loss of performance due to a worse channel estimation.

(26) A conservative value for ?T.sub.ch would be given by the differential delay between the slowest and the fastest mode when propagated through the few-mode fiber:

(27) $\begin{array}{cc}?T_{\mathrm{ch}}={?}^{\mathrm{slowest}}-{?}^{\mathrm{fastest}}=L.Math.\left(\frac{1}{v_g^{\mathrm{slowest}}}-\frac{1}{v_g^{\mathrm{fastest}}}\right)& 3\end{array}$

(28) Where v.sub.g.sup.slowest and v.sub.g.sup.fastest are the group velocities of the slowest and fastest modes, respectively.

(29) The length of the guard-band in FIG. 6 can guarantee that the whole set of training sequences transmitted in all the modes fit once parallel-to-serial operation is performed. Assuming N modes are transmitted, the time duration of the guard-band is given by:
T.sub.GB=N.Math.T.sub.TS4

(30) Equation 4 can be modified to consider any deviation when delaying the different mode-demultiplexed signals in point B of FIG. 5.

(31) Finally, the retrieval of the training sequences can be done through more than one LP0x mode. In this case, the coupling losses at point M (FIG. 5) can be reduced, as explained for the first prototype. However, in this case, using more than one LP0x mode has one more advantage, it also allows for the reduction of the time guard band. For instance, for 6 modes (LP01, LP02, LP11a, LP11b, LP21a, LP21b), the TS of the LP01, LP11a and LP11b can be transmitted back over LP01 and the TS of LP02, LP21a and LP21b can be transmitted over LP02. In this way, the T.sub.GB can be reduced by a factor of two. For more than 6 modes, similar reasoning applies.

(32) In order to test the configuration proposed in FIG. 5, the performance of a MDM system supporting 3 modes (LP01, LP11a, LP11b) and one 40 Gbps QPSK system per mode is modelled and simulated. The pre-equalization at the CO is achieved using a coherent receiver to estimate the downlink transfer matrix from the training sequences being retrieved just after the mode DEMUX. Furthermore, a simple direct detection ONU (without DSP) is used. The crosstalk introduced by the fibre is varied from ?50 dB/km to 0 dB/km, the mode non-selective MUX/DEMUX are considered to introduce an arbitrary high crosstalk value (full mixing), and the LP01 mode launch/select scheme is assumed to be ideal. The fibre is assumed to introduce a differential mode delay of 40 ps/km between LP01 and LP11a/b and 1 ps/km between LP11a and LP11b.

(33) FIG. 7 shows the Q-factor as a function of the fibre crosstalk, where the Q-factor of the worst mode is presented. The results show that for a fibre presenting a crosstalk strength as high as ?20 dB/km, the prototype proposed would be able to correctly pre-compensate the crosstalk for the full set of modes.

(34) The implementation of the architecture presented in FIG. 3 may use an electrical feedback loop circuit to command the electrical butterfly coefficients of the FIR filters that are used to provide pre-compensation. The implementation of the architecture in FIG. 5 may use an array of optical delay lines for the parallel-to-serial conversion of the transmitted downlink training sequences.

(35) The mode selective MUX/DEMUX can be implemented using phase plates in a free-space optics configuration as shown in FIG. 8.

(36) The embodiments discussed above can present several advantages. Both embodiments concentrate the signal processing effort in the central office, thereby avoiding the usage of digital signal processing in the ONUs. The embodiment that uses a mode selective MUX/DEMUX simplifies the OLT it simplifies the transmitter as only pilot tones are required, and it simplifies the receiver as only the usage of direct detection and simple electrical butterfly fir filters are required. The embodiment that uses a mode non-selective MUX/DEMUX allows the usage of lower cost optical components however it requires the usage of field detection techniques and digital signal processing.

(37) In an embodiment that uses the mode selective MUX/DEMUX, one pilot tone can be used per each pair of degenerate modes. Moreover, each pilot tone will have a different frequency such that when they are added after the DEMUX no interference takes place. The pilot tones can be used to update the FIR filter coefficients in two ways: Maximization the power of the retrieved/detected pilot tone is maximized when the pilot tone is launched in the ordinary mode and collected from the ordinary mode; Minimization: the power of the retrieved/detected pilot tone is minimized when the pilot tone is launched in the ordinary mode and collected from the extraordinary mode.

(38) The pilot tones may be low-frequency in-band tones or out-of-band high frequency tones. The pilot tones can be