Forward filter training scheme
A method includes receiving an input signal carrying transmitted values by a receiver, which includes a decision device that produces estimates of the transmitted values. The input signal is filtered using a forward filter to produce a forward-filtered signal, and the estimates of the transmitted values are filtered using a backward filter to produce a backward-filtered signal. The decision device is applied to a composite signal produced from the forward-filtered and backward-filtered signals, so as to produce the estimates of the transmitted values. An accuracy of the estimates of the transmitted values is assessed, and coefficients of the forward filter are adapted depending on the assessed accuracy.
1. A method, comprising:
receiving an input signal carrying transmitted values by a receiver, which includes a decision device that produces estimates of the transmitted values;
filtering the input signal using a forward filter to produce a forward-filtered signal, and filtering the estimates of the transmitted values using a backward filter to produce a backward-filtered signal;
applying the decision device to a composite signal produced from the forward-filtered and backward-filtered signals, so as to produce the estimates of the transmitted values;
assessing an accuracy of the estimates of the transmitted values produced by the decision device; and
adapting coefficients of the forward filter depending on the assessed accuracy, by generating a noise signal whose magnitude depends on the assessed accuracy of the estimates produced by the decision device, injecting the noise signal to an input of the forward filter, and adapting the coefficients while the forward filter operates on the noise signal.
2. The method according to claim 1 , wherein assessing the accuracy comprises computing a statistical measure of the accuracy of the estimates over multiple transmitted values, and wherein adapting the coefficients comprises adjusting the coefficients based on the computed statistical measure of the accuracy.
3. The method according to claim 1 , wherein adapting the coefficients comprises injecting the noise signal into a duplicate forward filter, adapting the coefficients of the duplicate forward filter responsively to the noise signal, and setting the coefficients of the forward filter to match the adapted coefficients of the duplicate forward filter.
4. The method according to claim 1 , wherein generating the noise signal comprises generating a digital Pseudo-Noise (PN) sequence.
5. The method according to claim 1 , wherein adapting the coefficients comprises adjusting the coefficients in a leaky Least Mean Square (LMS) adaptation process having a leakage factor that is selected depending on the assessed accuracy of the estimates.
6. The method according to claim 1 , wherein adapting the coefficients comprises initially setting the coefficients irrespective of the estimates, and subsequently adapting the coefficients so as to cause the forward filter to match the response of a communication channel over which the input signal is received.
7. The method according to claim 1 , wherein receiving the input signal comprises storing a first plurality of samples of the input signal and a second plurality of the estimates of the transmitted values in a memory, and wherein adapting the coefficients comprises processing the stored first and second pluralities.
8. The method according to claim 7 , wherein processing the stored first and second pluralities comprises applying a series of Least Mean Square (LMS) iterations to the stored first and second pluralities.
9. The method according to claim 1 , wherein assessing the accuracy and adapting the coefficients comprise applying a series of adaptation iterations, and adapting the coefficients in a given iteration assuming the accuracy of the estimates is improved relative to a previous iteration that precedes the given iteration in the series.
10. The method according to claim 1 , wherein the forward filter comprises an adaptive equalizer.
11. The method according to claim 1 , wherein the transmitted values in the input signal are encoded with an Error Correction Code (ECC), and wherein the decision device comprises a decoder of the ECC.
12. Apparatus, comprising:
a decision device, which is configured to produce estimates of transmitted values carried by a received input signal;
a forward filter, which is configured to filter the input signal to produce a forward-filtered signal;
a backward filter, which is configured to filter the estimates of the transmitted values to produce a backward-filtered signal; and
circuitry, which is configured to apply the decision device to a composite signal produced from the forward-filtered and backward-filtered signals so as to produce the estimates of the transmitted values, to assess an accuracy of the estimates of the transmitted values produced by the decision device, and to adapt coefficients of the forward filter depending on the assessed accuracy, by generating a noise signal whose magnitude depends on the assessed accuracy of the estimates produced by the decision device, injecting the noise signal to an input of the forward filter, and adapting the coefficients while the forward filter operates on the noise signal.
13. The apparatus according to claim 12 , wherein the circuitry is configured to assess the accuracy by computing a statistical measure of the accuracy of the estimates over multiple transmitted values, and to adapt the coefficients based on the computed statistical measure of the accuracy.
14. The apparatus according to claim 12 , and comprising a duplicate forward filter, wherein the circuitry is configured to inject the noise signal into the duplicate forward filter, to adapt the coefficients of the duplicate forward filter responsively to the noise signal, and to set the coefficients of the forward filter to match the adapted coefficients of the duplicate forward filter.
15. The apparatus according to claim 12 , wherein the circuitry is configured to generate the noise signal by generating a digital Pseudo-Noise (PN) sequence.
16. The apparatus according to claim 12 , wherein the circuitry is configured to adjust the coefficients in a leaky Least Mean Square (LMS) adaptation process having a leakage factor that is selected depending on the assessed accuracy of the estimates.
17. The apparatus according to claim 12 , wherein the circuitry is configured to initially set the coefficients irrespective of the estimates, and to subsequently adapt the coefficients so as to cause the forward filter to match the response of a communication channel over which the input signal is received.
18. The apparatus according to claim 12 , wherein the circuitry is configured to store a first plurality of samples of the input signal and a second plurality of the estimates of the transmitted values in a memory, and to adapt the coefficients by processing the stored first and second pluralities.
19. The apparatus according to claim 18 , wherein the circuitry is configured to apply a series of Least Mean Square (LMS) iterations to the stored first and second pluralities.
20. The apparatus according to claim 12 , wherein the circuitry is configured to apply a series of adaptation iterations, and to adapt the coefficients in a given iteration assuming the accuracy of the estimates is improved relative to a previous iteration that precedes the given iteration in the series.
21. The apparatus according to claim 12 , wherein the forward filter comprises an adaptive equalizer.
22. The apparatus according to claim 12 , wherein the transmitted values in the input signal are encoded with an Error Correction Code (ECC), and wherein the decision device comprises a decoder of the ECC.