Adaptive FIR filter and method

Abstract

Finite impulse response filters are commonly used in high speed data communications electronics for reducing error rates in multilevel symbol encoding schemes. Schemes such as pulse amplitude modulation and quadrature amplitude modulation may have higher error rates for symbols with low signal to noise ratios. By selectively updating the tap coefficients of the filter based on the symbols received, a more robust, accurate filter can be built.

Description

FIELD

The invention relates to finite impulse response filters, and, more specifically, adaptive finite impulse response filters.

BACKGROUND

Finite impulse response (FIR) filters are commonly used in high speed data communications electronics for reducing error rates in multilevel symbol encoding schemes. Schemes such as pulse amplitude modulation (PAM) and quadrature amplitude modulation (QAM) may have higher error rates for symbols with low signal to noise ratios. A symbol refers to a data pulse on a lane. The symbol may have several bits of information encoded within the voltage level and polarity (PAM for example). By using high levels of bit encoding in symbols, lower symbol rates can be used. Lower symbol rates are desirable because of their lower levels of noise.

FIG. 1 shows an example of a symbol pulse. The symbol pulse 102 shows the analog voltage vs. time response of the symbol being driven on the line. In an ideal world, the symbol pulse 102 would look like a square wave. Unfortunately, real pulses have profiles like that shown in FIG. 1. The line is driven so that at time t0 the amplitude of the pulse 102 is about h0. As the pulse 102 is driven back to zero, it still has an amplitude of h1 at time t1. The spacing of times t0 and t1 corresponds with the symbol rate. Thus, the receiver might confuse the pulse 102 amplitude at time t1 with a symbol meant to be valid at time t1, when it was simply the tail end of the pulse 102 meant to be valid at time t0.

FIG. 2 shows an example of two successive four level PAM symbols. Four level PAM usually has an encoding scheme with four different voltage levels: +3, +1 , −1, and −3 volts. This scheme gives two bits of encoding. Two successive symbols are shown with their possible waveforms for each pulse, one intended to be valid at time t0 and one at t1. The first symbol has possible pulses peaking at time t0 and has the following possibilities: a +3 pulse 202, a +1 pulse 204, a −1 pulse 206, or a −3 pulse 208. Similarly, the second symbol has possible pulses peaking at time t1 and has the following possibilities: a +3 pulse 210, a +1 pulse 212, a −1 pulse 214, or a −3 pulse 216. One can see that a +3 pulse 202 looks like +3 volts at time t0 and +1 volt at time t1.

FIG. 3 shows an example of a one tap FIR filter. The filter 302 provides a method of determining how much of first symbols tail needs to be removed from the apparent head of a second symbol. Data 303, which may be digital decoded, passes through a delay block 304. The delay block 304 delays the propagation of the data 303 for one period corresponding to the symbol rate. The data 303 then goes into the pulse block 306, where the sign of the data 303 is determined. It is also possible to determine the sign of the data 303 before the pulse block 306 and simply feed the sign of the data to the pulse block 306. The pulse block 306 also receives the sign of the error 308, which is determined by the difference between a symbol pulse and its decoded data at a given clock edge.

The output of a prior art pulse block 306 is the sign of the data 303 multiplied by the sign of the error 308. This output in the up/down pulse, and it is fed into an integrator 310. The integrator multiplies the up/down pulse by a small constant _— and adds that value to the previous filter coefficient to create a new coefficient. The following equation summarizes this:h_i+1=h_i+(_*sign(data)*sign(error)) Thus, the coefficient of the filter 302 constantly changes based on the sign of the error 308 and the sign of the data 303. The coefficient output of the integrator 310 is multiplied by the delayed data 303 at a multiplier 312. The result is a correction signal 314 that, when added to the symbol signal, ideally cancels any signal portions of previous symbols from a current symbol before the current symbol is decoded. Actual filters 302 may employ more than one tap in order to achieve the desired results.

FIG. 4 shows a chart of a prior art up/down pulse determination scheme. The sign of the data multiplied by the sign of the error determines the sign of the up/down pulse. Unfortunately, this prior art scheme does not work very well for noisy systems and symbols with small signals.

FIG. 5 shows an example of amplitude error and noise for symbols. The symbols 3, 1, −1, and −3 508, 506, 504, 502 respectively each have error distributions in their amplitudes. Also shown is a noise 510 distribution. For symbols with low signal to noise ratios, like symbol 1 506 and symbol −1 504, the noise 510 in the system may disrupt the accuracy of the up/down pulse. This, in turn, may erroneously affect the modification of filter coefficients.

Thus, there is a need for a FIR filter that accounts for low signal to noise ratio symbols when updating filter coefficients.

SUMMARY

This document describes a FIR filter where filter coefficients are selectively updated. This document also describes a method for selectively updating filter coefficients.

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1 shows an example of a prior art symbol pulse.

FIG. 2 shows an example of two successive prior art four level PAM symbols.

FIG. 3 shows an example of a prior art one tap FIR filter.

FIG. 4 shows a chart of a prior art up/down pulse determination scheme.

FIG. 5 shows an example of prior art amplitude error and noise for symbols.

DESCRIPTION

FIR filters are used in many types circuits. Examples include decision-feedback equalizers (DFE), echo cancellers (EC), feed forward equalizers (FFE), automatic gain control (AGC), and adaptive reference control (ARC). By not updating the filter coefficients for symbols with low signal to noise ratios, a more robust, less noise sensitive filter can be created.

FIG. 6 shows an example of a selective up/down pulse scheme for four level PAM. With four level PAM, the +3, −3 symbols have high signal to noise ratios, while the +1, −1 symbols have low signal to noise ratios. By changing the possible output of a pulse block (from −1, 1 in the prior art) to −1, 0, 1, the filter is allowed to keep its coefficients constant for symbols with low signal to noise ratios. The scheme shown in FIG. 6 outputs the sign(data)*sign(error) for symbols of +3 and −3, but the scheme also has an output of zero for symbols of +1 and −1. An output of zero means that an integrator will keep a tap coefficient constant for that symbol. The implementation of the modified up/down pulse scheme may be, among others, a lookup table or simple logical circuit. The implementation looks at both the sign and magnitude of the symbol.

The method of selectively updating filter coefficients may be applied to any encoding scheme. For example, with eight level PAM, one may choose to only update the coefficients for symbols with the largest absolute values (for example, +7 and −7, where the symbols are −7, −5, −3, −1, 1, 3, 5, 7). In another example, with 12 level PAM, one may choose to only update the coefficients for symbols with the largest absolute values (for example, +11 and −11, where the symbols are −11, −9, −7, −5, −3, −1, 1, 3, 5, 7, 9, 11). Alternately, one may choose to update the coefficients only for symbols with absolute signals in the upper half of possibilities. Alternately, one may choose to update the coefficients for all symbols except those with the smallest absolute signals (for example, update for all but 1 and −1 in 12-PAM). The scheme may also be used for quadrature encoded symbols. The symbols with low signal to noise ratios simply do not trigger an update of the filter coefficients.

FIG. 7 shows an example of a three tap DFE filter. The DFE filter 700 includes three taps 702a, 702b, 702c. The taps 702a, 702b, 702c each comprise a FIR filter with a selective coefficient updating scheme. The output of the taps is added to an incoming signal 704 at a summing node 706. The output of the summing node 706 ideally has a level corresponding to the current symbol. The decision circuit 708 decodes the signal into data 710. The difference between the data 710 and the symbol is determined at a subtracting node 712. This difference comprises the error. Feeding the error through a slicer 714 results in the sign of the error, which is fed to each tap 702a, 702b, 702c of the filter 700. The table blocks noted in each tap 702a, 702b, 702c comprise the up/down pulse blocks that select an up, down, or zero pulse for the integrator. The table blocks may be lookup tables, logic, or even an adaptive algorithm that determines for which symbols the coefficients will be updated. The schemes used in each tap may not necessarily be the same.

It will be apparent to one skilled in the art that the described embodiments may be altered in many ways without departing from the spirit and scope of the invention. Accordingly, the scope of the invention should be determined by the following claims and their equivalents.

Claims

1. A finite impulse response filter comprising: at least one tap, each tap comprising: an integrator; a delay element; and a multiplier, where the multiplier is connected to the integrator and the delay element, where output of the integrator is a tap coefficient, and where the tap coefficient is not changed for data values with low signal to noise ratios.

2. The method of claim 1, further comprising a lookup table, where the lookup table is connected to the integrator.

3. The filter of claim 1, wherein the filter reads pulse amplitude modulated encoded signals.

4. The filter of claim 3, wherein the signal comprises a four level pulse amplitude modulated signal, wherein each tap coefficient is updated for symbols of +3 and −3, and wherein each tap coefficient is not updated for symbols of +1 and −1.

5. The filter of claim 3, wherein the signal comprises a eight level pulse amplitude modulated signal, wherein each tap coefficient is updated only for symbols of +7 and −7, and wherein each tap coefficient is not updated for symbols other than 7 and −7.

6. The filter of claim 3, wherein the signal comprises a twelve level pulse amplitude modulated signal, wherein each tap coefficient is updated only for symbols of +11 and −11, and wherein each tap coefficient is not updated for symbols other than 11 and −11.

7. The filter of claim 1, wherein the filter reads quadrature amplitude modulated encoded signals.

8. A method of adjusting coefficients of a finite impulse response filter comprising: monitoring data; monitoring error; and adjusting the filter coefficients based on the data and error, where the coefficients are selectively adjusted.

9. The method of claim 8, wherein the filter handles signals with more than two levels of encoding.

10. The method of claim 9, wherein the encoding comprises pulse amplitude modulation.

11. The method of claim 9, wherein the encoding comprises quadrature amplitude modulation.

12. The method of claim 8, wherein the coefficients are adjusted for a first data range with a high signal to noise ratio, and where the coefficients are not adjusted for a second data range with a low signal to noise ratio.

13. The method of claim 8, wherein the coefficients are adjusted for a larger half of possible data values, and where the coefficients are not adjusted for a smaller half of possible data values.

14. A method of adjusting coefficients of a finite impulse response filter comprising: monitoring data; monitoring error; and adjusting the filter coefficients based on the data and error, where the signal is a four level pulse amplitude modulation signal, where the coefficients are adjusted for symbols of +3 and −3, and where the coefficients are not adjusted for symbols of +1 and −1.