Configurable decimation, interpolation and shaping filters for wideband receive and transmit chains. The arithmetic is textbook; the engineering is in spending the fewest multipliers and the least memory for a given stopband, and in sizing precision per stage rather than uniformly across the datapath.
A filter chain sits between a converter running far faster than the signal needs and a datapath that cannot afford to. Getting from one rate to the other cheaply is most of the work in a wideband front end.
Doing it in one stage is almost always wrong. A single filter decimating by a large factor needs an enormous number of taps to hold its transition band; a cascade — a cheap CIC to take the bulk of the rate down, half-band stages to halve efficiently, then a shaping FIR at the low rate where taps are affordable — achieves the same response for a fraction of the resource.
Precision is the other lever. Applying a uniform word length across a datapath wastes bits at the wide end and starves the narrow end. Sizing each stage against the dynamic range it actually carries recovers significant DSP and memory capacity with no measurable performance loss, which is frequently what allows a design to close in the device that was budgeted.
| Parameter | Detail |
|---|---|
| Filter types | FIR, CIC with droop compensation, half-band, and polyphase resampling |
| Rate change | Integer decimation and interpolation; fractional resampling where rates are not commensurate |
| Architecture | Cascaded stages, so the expensive taps run at the lowest rate that will carry them |
| Symmetry exploitation | Folded FIR structures, halving multiplier count on linear-phase responses |
| Precision | Per-stage word length analysis rather than a uniform datapath width |
| Coefficients | Fixed or runtime-reloadable, with sets switchable per operating mode |
| Response control | Passband ripple, stopband attenuation and transition width traded explicitly against resource |
| Target | FPGA fabric or ASIC; DSP slice mapping where the device provides it |
Maturity — silicon-proven, FPGA-validated or RTL stage — is confirmed at enquiry for the specific configuration you need, rather than claimed generically here.
Send the target node, the interface requirements and the integration context. If this is not the right fit, that will be said early rather than discovered at integration.
A single filter decimating by a large factor needs an enormous number of taps to hold its transition band, and they all run at the input rate. A cascade — CIC for the bulk rate reduction, half-band stages to halve efficiently, then a shaping FIR at the low rate — achieves the same response for a fraction of the multipliers and memory.
A CIC filter is extremely cheap because it uses no multipliers, but its passband is not flat — it sags toward the band edge. That droop is corrected by a compensating filter in a later, lower-rate stage, which is far cheaper than avoiding the droop in the first place.
Enough to change device selection on real designs. A uniform word length wastes bits where the dynamic range is narrow and starves the datapath where it is wide; sizing each stage to what it carries recovers significant DSP and memory with no measurable performance loss.
Either. Fixed coefficients are cheaper; reloadable sets allow bandwidth and response to change with operating mode, which matters where one radio serves several waveforms or channel plans.
Splitting a wideband capture into channels.
PRODUCTThe datapath building block.
TECHNOLOGYWhat sits downstream of the filter chain.
PRODUCTWhere these filters are deployed.
CASE STUDYPrecision analysis that closed the device.
CAPABILITYClosing a datapath in the budgeted part.
Tell us the specification, the constraint and the deadline. Programmes that cross silicon, radio, embedded and AI are where Faststream is strongest.