Digital filters

• US 4,771,396 A
• Filed: 03/14/1985
• Issued: 09/13/1988
• Est. Priority Date: 03/16/1984
• Status: Expired due to Fees

• Current Assignee: British Telecommunications PLC (BT Group PLC)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: South, Colin R., Lewis, Alwyn V.
• Primary Examiner(s): Harkcom, Gary V.
• Assistant Examiner(s): Shaw, Dale M.
• Application Number:
  US06/711,686
• Time in Patent Office: 1,279 Days
• Field of Search: 364/724
• US Class Current: 708/319
• CPC Class Codes: H03H 2021/0089 System identification, i.e. modeling
• CPC Class Codes: H03H 21/0012 Digital adaptive filters
• CPC Class Codes: H04B 3/23 using a replica of transmitted signal in the time domain, e.g. echo cancellers
• CPC Class Codes: H04B 3/238 using initial training sequence

Abstract

A transversal finite impulse response, adaptive digital filter comprises three stages (10, 11, 12). The stage (10) is a memory stage which stores input samples x and coefficients. The second stage (11) is an arithmetic stage which forms convolution products and accumulates the product to provide an output. The third stage (12) forms update values for updating the coefficients according to a predetermined algorithm. The stages are coupled by buses (14, 15) and data transfer between the stages via the buses is controlled by a control unit (18). The filter operates so that convolution products are formed during a first part of each sample period and update values are formed during a second part of each sample period for use in the subsequent sample period.

Claims

1. An adaptive digital filter comprising means (10) for receiving and storing input samples (x) during recurring sample periods, means (11) for forming, during each sample period from said input samples and filter coefficients, a plurality of convolution products and for accumulating said plurality of products an output signal (r), and means (12) for adjusting said filter coefficients according to a predetermined algorithm, the forming means forming the said convolution products during a first part of each sample period and the adjusting means adjusting said filter coefficients, in response to said convolution products formed during said sample period first part, at least in part during a second part of each sample period which is subsequent to said sample period first part.

2. A digital filter as claimed in claim 1 wherein:

said receiving and storing means includes a memory stage (10) for storing input samples (x) and the filter coefficients (h),

said forming and accumulating means includes an arithmetic stage (11) for forming convolution products of the input samples and the said coefficients and accumulating the products to provide an output signal (r), and

said adjusting means includes means (12) for forming an update signal (h) adapting the filter coefficients according to a predetermined algorithm.

3. A digital filter as claimed in claim 2 wherein said memory stage (10) comprises a first section (30) for storing values of input samples (x), a second section (part of 33) for storing values of filter coefficients (ha) which can be adapted, and a third section (part of 33) for storing values of filter coefficient (hs) which are not adaptable.

4. A digital filter as claimed in claim 3 including means for transferring the contents of the second memory section to the third memory section.

5. A digital filter as claimed in claim 3 or 4 wherein the arithmetic stage includes means (35) for forming a first set of convolution products between values of the input signal (x) and filter coefficients (ha) from the second memory section, means (35) for forming a second set of convolution products between values of the input signal (x) and filter coefficients (hs) from the third memory section, and means (136) for accumulating said first set of convolution products to form a first output signal and for accumulating said second set of convolution products to form a second output signal.

6. A digital filter as claimed in any one of claims 2 to 3 including bus means coupling said stages to permit data transfer between them, and control means (18,20) for providing timing signals to control operation of said stages and transfer of data between said stages.

7. A digital filter as claimed in claim 6 wherein the input samples (x) and coefficient values (h) are multiplexed onto the bus means (14,15) for transfer from the memory stage (10) to the arithmetic stage (11) under the control of clock signals from said control means (18,20).

8. A filter as claimed in any one of claims 2 to 3 wherein:

said adjusting means updates values for the filter coefficients by adding said preset coefficients to a correction factor equal to the product of (a) a scaling factor, (b) an error factor representing the difference between the filter output and a desired signal, and (c) a respective one of the last W input samples, where W is the number of coefficients; and

said forming and accumulating means performs a multiplication operation on the input samples to obtain said convolution products and adds said convolution products to said coefficients during the first part of the sample period, and said adjusting means multiplies the scaling factor by the error factor during the second part of the preceding sample period.

9. A digital filter as claimed in claim 8 wherein said arithmetic stage includes means (41,43) arranged during each sample period to form from sample values x values of x², accumulated over a predetermined number of samples, which are fed to the third stage (12) for use in forming said update values, and wherein the said predetermined algorithm for updating the filter coefficients H can be expressed as

10. A filter according to claim 9 in which,

(a)in the second part of the (n-1)th sample period, the scaling factor f(2μ

) is derived from the accumulated value of x² ;

a new sample (x) is received, and the square of this new value is added to the accumulated value of x², and the error between the filter output and the derived value is obtained and its product with f(2μ

) obtained,

(b) in the first part of the nth sample period, a plurality W of process cycles occur in each of which a respective one of the stored sample values x, x_n-1 . . . X_n-w-1 is multiplied by the said product and added to the respective one of the filter coefficients; and

the updated value of the coefficient is multiplied by the corresponding one of the stored sample values and accumulated, whereby at the end of the first part of the nth sample period an accumulated filter output is obtained using the error obtained from the immediately preceding coefficient update and

(c) in the second part of the nth sample period the sample value x_n-W is squared and subtracted from the accumulated x² value.

11. A digital filter as claimed in any one of claims 2 to 3 wherein the step size provided by said third stage can be adjusted.

12. A loudspeaking telephone incorporating a digital filter as claimed in any one of claims 1 to 3.

13. A digital filter which comprises a plurality of filters according to claim 1 arranged in cascade.

14. An adaptive digital filter comprising a memory stage for storing input samples and filter coefficients, an arithmetic stage connected to said memory stage for forming convolution products of said input samples and filter coefficients and for accumulating said products to provide an output signal during a predetermined sampling period, a third stage arranged to form an update signal for adapting said filter coefficients according to a predetermined algorithm substantially during said same predetermined sampling period, bus means coupling said stages to permit data transfer therebetween, and control means for providing timing signals to control operation of said stages and transfer of data between said stages.

15. An adaptive digital filter comprising:

means for sampling an input signal during recurring sampling periods each including a first portion and a second portion subsequent to said first portion;

means connected to said sampling means for forming convolution products during said sampling period first portions in response to said sampled input signal and preset filter coefficients, and for generating a filter output signal in response to said convolution products; and

means connected to said forming means for adjusting said preset filter coefficients during said sampling period second portions,

said adjusting means adjusting, substantially during the second portion of a given sampling period, said coefficients in response to convolution products formed from input signal samples obtained during the first portion of said same given sampling period, said adjusting means making said adjusted coefficients available at least by the time said first portion of the sampling period immediately succeeding said given sampling period occurs.

16. In an adaptive digital filter of the type which receives and samples a digital input signal, multiplies said input signal samples by filter coefficient values to form convolution products, and accumulates the convolution products to produce a filtered output, an improvement comprising:

(a) forming convolution products during an initial, burst portion of a first sampling period in response to a digital input signal sample associated with said first sampling period and in response to said filter coefficient values;

(b) adjusting said filter coefficient values in response to said convolution products formed by said forming step during said first sampling period burst portion substantially during a pause portion of said first sampling period subsequent to said burst portion; and

(c) forming convolution products during an initial, burst portion of a second sampling period immediately succeeding said first sampling period in response to a digital input signal sample associated with said second sampling period and also in response to said filter coefficient values adjusted by said adjusting step (b).

17. In an adaptive digital filter of the type which receives and samples a digital input signal, multiplies said input signal samples by preset filter coefficient values to form convolution products, and accumulates the convolution products to produce a filtered output, an improvement comprising:

means for forming convolution products during an initial, burst portion of a first sampling period in response to a digital input signal sample associated with said first sampling period and also in response to said filter coefficient values; and

means for adjusting said filter coefficient values, in response to said convolution products formed by said forming means, substantially during a pause portion of said first sample period subsequent to said burst portion,

wherein said forming means also forms convolution products during an initial, burst portion of a second sampling period immediately following said first sampling period in response to a digital input signal sample associated with said second sampling period and also in response to said filter coefficients adjusted by said adjusting means during said first sampling period pause portion.