System and method for filtering random noise using data compression

• US 5,400,371 A
• Filed: 03/26/1993
• Issued: 03/21/1995
• Est. Priority Date: 03/26/1993
• Status: Expired due to Term

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Hewlett-Packard Company (HP Inc.)
• Inventors: Natarajan, Balas K.
• Primary Examiner(s): Chin, Stephen
• Assistant Examiner(s): VO, DON NGUYEN
• Application Number:
  US08/037,774
• Time in Patent Office: 725 Days
• Field of Search: 375/99, 375/103, 375/122, 364/724.01, 364/724.19, 348/384, 348/438, 348/607, 348/618, 382/56
• US Class Current: 375/240
• CPC Class Codes: H04B 1/62 for providing a predistortion of the signal in the transmitter and corresponding correction in the receiver, e.g. for improving the signal/noise ratio

Abstract

In a system in which a non-random, noise-free signal is disturbed by random noise, one first determines a way to measure the difference between two signals, and one also selects a measure of complexity for signals. Based on a series of discrete values of the noisy signal, a compression processor generates a series of compressed signals representing the noisy signal, each within a corresponding error or loss tolerance of the discrete values of the noisy signal. An optimization processor applies the various loss tolerance values to the compression processor and then evaluates the relative complexity of the corresponding compressed signals. The optimization processor then determines an optimal knee point loss tolerance, below which the complexity of the compressed signals rises rapidly. For continued filtering of the noisy signal, the compression processor compresses the noisy signal using the optimal knee point loss tolerance. The optimally compressed signals are then passed to a decompression processor, which generates a filtered signal corresponding to the noisy signal. In certain embodiments, the optimization processor also controls the sampling rate of the noisy signal; this makes possible a reduction in the residual loss of the filtered signal.

Claims

1. A system for filtering a noisy signal, which includes a desired signal corrupted by random noise, comprising:

A. a source of discrete samples of the noisy signal;

B. a compression processor having, as an uncompressed input, the discrete samples of the noisy signal, and having as a compressed output a compressed representation of an approximation signal that differs from the uncompressed input by no more than an adjustable error tolerance, which is a predetermined function of a predetermined error metric and is smaller than the uncompressed input in a predetermined complexity measure;

C. a decompression processor having the compressed representation of the approximation signal as a compressed input and having the approximation signal as a decompressed output; and

D. an optimization processor that has;

1. an optimization input connected to the compressed output of the compression processor to receive the compressed representation of the approximation signal; and

2. a tolerance output connected to the compression processor and conveying the adjustable error tolerance of the compression processor.

2. A system as defined in claim 1, further including a memory unit containing the discrete samples of the noisy signal.

3. A system as defined in claim 1, in which:

the source of discrete samples is a sampling circuit having as a sampling input the noisy signal and having as a sampled output the discrete values; and

the optimization processor is connected to the sampling circuit for applying a sampling rate signal to the sampling circuit.

4. A system as defined in claim 1, in which the compression processor, the decompression processor, and the optimization processor are a single unit.

5. A system for filtering a noisy signal, which includes a desired signal corrupted by random noise, comprising:

A. discretizing means for generating discrete samples of the noisy signal;

B. compression means for receiving the discrete samples of the noisy signal as an uncompressed input, for generating a compressed representation of an approximation signal that differs from the noisy signal by no more than an adjustable error tolerance, which is a predetermined function of a predetermined error metric and is smaller than the uncompressed input in a predetermined complexity measure;

C. decompression means for receiving the compressed representation of the approximation signal, and for generating the approximation signal as a decompressed output;

D. optimization means for repeatedly applying to the compression means different values of the adjustable error tolerance, for receiving the corresponding compressed representations of the approximation signals, for computing in the predetermined complexity measure the complexity value of the compressed representation obtained for each of the error tolerance values, and for selecting a knee-point tolerance value.

6. A system as defined in claim 5, further including memory means for storing the discrete samples of the noisy signal.

7. A system as defined in claim 5, in which:

the discretizing means is a sampling means for sampling the noisy signal and for generating the discrete samples; and

the optimization means is connected to the sampling means for varying the sampling rate of the sampling means.

8. A method for filtering a noisy signal, which includes a desired signal corrupted by random noise, comprising the following steps:

A. selecting an error metric and a complexity measure;

B. for each of a plurality of error tolerance values, repeating steps C-E as follows;

C. selecting one of the error tolerance values;

D. in a compression processor, generating a compressed data signal that differs from the noisy signal by no more than the selected error tolerance value according to the selected error metric;

E. in an optimization processor, for each generated compressed data signal, calculating a corresponding complexity value using the selected complexity measure;

F. determining a knee-point error tolerance value as the value at which the complexity values have maximum acceleration with respect to the error tolerance value;

G. setting a running error tolerance rate for the compression processor equal to the knee-point error tolerance value; and

H. filtering the noisy signal by compressing it in the compression processor with the running error tolerance rate, and decompressing it in a decompression processor.

9. A method as defined in claim 8, further including the following steps:

A. sampling the noisy signal to generate a plurality of noisy discrete values of the noisy signal;

B. storing the plurality of noisy discrete values in a memory circuit;

C. in which the step of generating the compressed data signal in the compression processor includes applying the plurality of stored, noisy discrete values of the noisy signal as input signals to the compression processor.

10. A method as defined in claim 8, in which the maximum acceleration of the complexity values is determined by generating a bilinear approximation of the relationship between the complexity values and the error tolerance values and picking as the knee-point error tolerance value the error tolerance value corresponding to a breakpoint of the bilinear approximation.

11. A method as defined in claim 8, including the following additional steps:

A. generating discrete values of the noisy signal;

B. generating in the optimization processor a plurality of different sampling rates of the noisy signal and for each of the plurality of sampling rates, repeating step C as follows;

C. for each of the plurality of error tolerance values, repeating steps

     1)-3) as follows;

1) selecting one of the error tolerance values;

2) in the compression processor, generating the compressed data signal that differs from the noisy signal by no more than the selected error tolerance value according to a predetermined error metric; and

3) in the optimization processor, for each generated compressed data signal, calculating a corresponding complexity value using a predetermined complexity metric;

D. determining a knee-point error tolerance value as the error tolerance value for which the complexity values for the different sampling rates diverge by more than a predetermined threshold value.