Polynomial Structure
Polynomial interpolation structures evaluate continuously variable signal delays without recalculating digital filter coefficients for every delay value. A farrow filter implements fractional delay by decomposing filter impulse responses into a set of fixed FIR base filters multiplied by powers of a fractional delay parameter. Data acquisition systems in digital oscilloscopes and software-defined radio receivers use this structure for arbitrary sample rate conversion.
Finite coefficient word length in the polynomial multipliers introduces quantization noise that degrades dynamic range.
Interpolation Architecture
Continuous delay adjustment allows real-time tuning of fractional sample timing in high-speed digital receiver channels. Within a farrow filter, coefficient matrices remain constant while a single control variable alters the effective delay between sampling instants. Lagrange or B-spline polynomials form the mathematical foundation for the sub-filter coefficients, balancing passband flatness against computational complexity.
Third-order polynomial configurations achieve sufficient attenuation of image spectra for mid-range instrumentation while keeping multiplication steps low. Phase linearity across the signal band prevents dispersion of transient sensor waveforms during fractional resampling operations. Calibration procedures verify magnitude response flatness against theoretical polynomial curves to ensure conversion fidelity.
Quantization Error
Fixed-point numerical representations in field-programmable gate arrays truncate sub-filter coefficients and internal accumulator outputs. These arithmetic truncation steps introduce spurious spectral tones that reduce signal to noise ratio. Coefficient precision of sixteen bits typically keeps idle channel noise below transducer noise floors in sensor interfaces.
Operational clock jitter in the delay control register causes timing noise that degrades high-frequency signal reconstruction.
Operational Envelope
Bandwidth constraints restrict maximum input frequencies to prevent aliasing during fractional delay adjustments. Sensor signals exceeding half the base sampling rate incur edge distortion.