Data sample compression and decompression using randomized quantization bins
AssigneeAmbarella, Inc.
InventorElliot N. Linzer
Life of this patent
Estimating the term…
Abstract
A method for compressing data samples using randomized quantization bins is disclosed. The method generally comprises steps (A) to (C). Step (A) may generate a size signal having a quantization step size corresponding to the data samples in an input signal using a compression circuit. The data samples generally reside in a range of numbers. Step (B) may generate the randomized quantization bins in the range based on a pseudorandom process. Step (C) may generate a plurality of quantized samples in a compressed signal by quantizing the data samples based on the randomized quantization bins.
Claims19
A method for compressing data samples, comprising the steps of:
- (A) generating a current one or more of a plurality of quantization step sizes corresponding to said data samples of an input signal;
- (B) generating a plurality of values based on both (i) a pseudorandom process and (ii) a plurality of positions of said data samples;
- (C) generating a plurality of intermediate samples by adjusting said data samples based on said values, wherein said pseudorandom process varies said values in response to said current quantization step size;
- and
- (D) generating a plurality of quantized samples in a compressed signal by quantizing said intermediate samples based on said current quantization step size.
An apparatus comprising:
- a first circuit configured to generate a current one or more of a plurality of quantization step sizes corresponding to a plurality of data samples of an input signal;
- and
- a second circuit configured to (i) generate a plurality of values based on both (a) a pseudorandom process and (b) a plurality of positions of said data samples, (ii) generate a plurality of intermediate samples by adjusting said data samples based on said values and (iii) generate a plurality of quantized samples in a compressed signal by quantizing said intermediate samples based on said current quantization step size, wherein said pseudorandom process varies said values in response to said current quantization step size.
A method for compressing, comprising the steps of:
- (A) generating a plurality of data samples by one of (i) a frequency-domain transformation and (ii) differential pulse code modulation of an input signal;
- (B) generating a plurality of values based on a pseudorandom process;
- (C) generating a plurality of intermediate samples by adjusting said data samples based on said values, wherein said pseudorandom process is based on at least one of (i) a temporal position, (ii) a spatial position and (iii) a frequency-domain position of said data samples;
- and
- (D) generating a plurality of quantized samples in a compressed signal by quantizing said intermediate samples.
Dependencies are inferred from the claim text.