Browse Prior Art Database

Method for a fast architecture for FIR implementation and LMS updates

IP.com Disclosure Number: IPCOM000009603D
Publication Date: 2002-Sep-04
Document File: 3 page(s) / 97K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for a fast architecture for first impulse response (FIR) implementation and least mean squares (LMS) updates. Benefits include improved performance.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 50% of the total text.

Method for a fast architecture for FIR implementation and LMS updates

Disclosed is a method for a fast architecture for first impulse response (FIR) implementation and least mean squares (LMS) updates. Benefits include improved performance.

Background

      The conventional LMS-update algorithm is yk(n) = yk(n-1) + me(n-1)x(n-k) for all FIR coefficients, k = 0, 1, … N-1, and for all time, n = -¥ to +¥.

              Conventionally, the FIR filter output at time n, z(n) is obtained by finding the convolution sum, z(n) = Syn(k). x(n-k), k goes from 0, 1, .. N-1. After the convolution sum for time, j, is completed, the LMS updates are obtained on the filter coefficients in a separate loop. The order of the tasks in the conventional algorithm, at time n, is:

1.           FIR filtering, for time n. Takes N cycles.

2.           Calculate the value me(n-1), for time n. Takes 1 cycle.

3.           LMS update for time n. Takes 2N cycles.

              Some processors with parallel arithmetic logical units (ALUs) can execute step 3 in N machine cycles.

Description

              The disclosed method is a fast architecture for FIR implementation and LMS updates. For example, an N-tap FIR filter is comprised of the following values:

•             x(n) is a sequence of input data samples, where in general, n = -¥ to +¥

•             yn(k) is the k-th filter coefficient of the filter at time instance n. Note that k = 0, 1, … N-1

•             z(n) = Syn(k). x(n-k), the convolution sum or output of the LMS FIR filter at iteration n

•             e(n) = z(n) – z’(n), the difference between the desired and actual filter output

•             m is the training coefficient

 

              Using the disclosed method, the following steps are performed:

1.           Calculate the value me(n-1), for time (n-1). Takes 1 cycle

2.           LMS update, for time (n-1)

3.           FIR filtering, for time n

              An algorithm (engine) performs the processing (see Figure 1). For every time instance, designated as n, th...