Fir filter with quantized coefficients and coefficient quantization method

A method of encoding a series of n-bit digital words, comprising:

• selecting a plurality of bits within each n-bit word, at least one of said selected bits being an active bit if the n-bit word includes at least some active bits,
• generating an m-bit digital word that represents said selected bits, where m is less than n,
• generating a shift word that represents the order of said selected bits within said n-bit word, said m-bit word when shifted by the value of said shift word to the scale of said n-bit word differing from said n-bit word by a quantization error, and
• feeding back said quantization error and combining it with the next n-bit word.

A finite impulse response (FIR) filter system, comprising:

• means for providing n-bit FIR filter coefficients,
• means for operating upon said n-bit coefficients to reduce them to lower order coefficients comprising combinations of respective m-bit quantized words and shift words, wherein m is less than n and said shift words relate said m-bit words to a binary order within their respective n-bit coefficients,
• means for receiving input data words, and
• multiplier means connected to convolve said data words with said lower order coefficients to yield a filtered output.

A digital-to-analog converter (DAC), comprising:

• means for increasing the sample rate of an input digital data signal,
• an interpolation finite impulse response (FIR) filter system connected to filter out images of the data signal produced by said sample rate increasing means,
• a hardware sigma-delta modulator connected to quantize the output of said FIR filter system to a series of + and - pulses,
• an integrator for integrating the output of said hardware sigma-delta modulator into an analog signal, and
• a low pass analog filter connected to filter the output of said integrator,
• wherein said FIR filter system comprises:
• means for providing n-bit FIR filter coefficients,
• a software sigma-delta modulator programmed to quantize said n-bit filter coefficients to respective m-bit quantized coefficients, where m is less than n, plus respective quantization error terms, and to shape said quantization error terms to a higher frequency range that is substantially outside the pass band of said low pass filter, and
• hardware multiplier means connected to convolve said input data signal with said quantized coefficients to produce a filtered output.

A method of quantizing a series of multi-bit digital words, comprising:

• providing a quantization function,
• sensing a characteristic of said multi-bit words,
• dynamically adjusting the quantization range of said quantization function for said words in accordance with their sensed characteristics, and
• operating upon said words with their respective adjusted quantization functions.