A new technique for sampling and reconstructing signals has been developed, allowing for the direct sampling of bandwidths much larger than the Nyquist bandwidth. This method, which can produce useful spectral measurements under specific conditions and applications, has been demonstrated in a built system. The system uses analog-to-digital converters (ADCs) sampling at 3 GHz, followed by DSP techniques to reconstruct an impressive 30 GHz of bandwidth—20 times the Nyquist bandwidth.

This guide was created with the help of AI, based on the presentation's transcript. Its goal is to give you useful context and background so you can get the most out of the session.

What this presentation is about and why it matters

This talk demonstrates a practical technique — called multi‑Nyquist sampling and reconstruction — that lets you measure and reconstruct spectral content well beyond the instantaneous Nyquist bandwidth of a single ADC. Instead of using an extremely fast (and power‑hungry / expensive) ADC to capture, say, 30 GHz of spectrum, the method uses multiple slower ADCs (clocked coherently but at slightly different rates) plus FPGA DSP (polyphase filter banks, aligned FFTs and cross‑correlation) to recover useful spectral estimates across many Nyquist zones.

Why this matters: many real systems (EMC test rigs, wideband receivers, radio‑astronomy arrays, and low‑power distributed sensors) need wide total coverage but cannot afford the power, cost or complexity of a single ultra‑fast sampler. Multi‑Nyquist sampling provides a trade: intentional aliasing (a form of controlled compression) followed by DSP reconstruction. The speaker shows a working demo: ADCs sampling near 3 GHz used to reconstruct roughly 30 GHz of spectrum — illustrating both feasibility and practical engineering choices.

Who will benefit the most from this presentation

• FPGA and DSP engineers interested in wideband spectrum capture without ultra-high‑rate ADCs
• RF and analog front‑end designers curious about front-end bandwidth tradeoffs and clocking requirements
• EMC test engineers and spectrum-monitoring practitioners looking for new monitoring architectures
• Students and researchers in sampling theory, software radio, and radio astronomy who want a practical demo of advanced sampling strategies
• System architects exploring low‑power, many‑node sensing (for example, remote arrays where telemetry and power are constrained)

What you need to know

Read this section before the talk to get the most out of the live demo and the FPGA/DSP descriptions.

• Nyquist sampling theorem (practical form): for a real signal sampled at rate $f_s$, the standard Nyquist (positive) band extends to $f_{Nyq}=f_s/2$. If content exists above that band, it will alias into the observed baseband unless filtered.
• Aliasing and Nyquist zones: spectral replicas of the input appear when sampling; for the listener, think of the spectrum as folding every Nyquist band boundary. In this talk the presenter uses multiple Nyquist zones deliberately rather than blocking them with a front‑end filter.
• Phase coherence between ADCs: to combine samples from different converters coherently you must share a stable reference clock (so samples are phase‑coherent). Coherent combination makes true signals add while out‑of‑band noise tends to average out.
• Ratio of sampling rates and bin alignment: the trick uses two sampling rates with a rational ratio (example in the talk: 21:20). Choosing FFT/DFT sizes proportional to each sampler’s rate aligns frequency bins so bins from different converters correspond to the same true frequency.
• Polyphase filter banks + FFTs: polyphase PFBs are used as the first stage in the FPGA to convert wide, multi‑lane ADC outputs into channelized single sample streams. A subsequent transform (e.g. 1024‑point) gives per‑bin spectra at a convenient resolution.
• Prime‑factor and Winograd algorithms: efficient DFT factorizations (Good‑Thomas, Winograd) reduce FPGA work when FFT sizes factor into small primes. The talk discusses choosing filter bank sizes with favorable prime factors so you can implement efficient transforms without extra twiddle rotation.
• Cross‑correlation / accumulation: the reconstruction is performed essentially by cross‑multiplying and integrating coherently across the aligned bins. More integrations reduce noise at the cost of throughput and latency.
• Practical limits: front‑end analog bandwidth, clock phase noise, ADC aperture jitter, and losses in cabling/components all limit how far you can push multi‑Nyquist reconstruction. The talk discusses these tradeoffs and practical FPGA layout choices.

Glossary

• Nyquist band / Nyquist zone — the frequency range up to $f_s/2$ for a sampler running at $f_s$; higher zones refer to successive folded bands.
• Aliasing — folding of spectral content into the sampled band when input frequencies exceed the Nyquist limit.
• ADC aperture jitter — timing uncertainty of the ADC’s sample instant; it causes phase noise at high input frequencies and limits SNR.
• Polyphase filter bank (PFB) — a channelizer implemented as windowed, decimated FIR stages feeding an FFT; useful to split wideband data into many narrow channels efficiently.
• FFT / bin width — discrete Fourier transform size sets spectral resolution; when bins are aligned across channels, corresponding bins map to the same true frequency.
• Phase coherence — fixed phase relationship between sample streams (usually by common clock/reference), enabling coherent addition and cross‑correlation.
• Good‑Thomas (prime‑factor) algorithm — a DFT factorization method that avoids twiddle factors when DFT sizes are relatively prime factors.
• Winograd Fourier Transform — an efficient small‑size DFT implementation that lowers multiplication count for specific composite sizes.
• Cross‑correlation / integration — multiplying one spectrum by the complex conjugate of another and summing over time to improve SNR of coherent components.
• Controlled aliasing (signal compression) — intentionally allowing many frequency bands to fold into a smaller observed band, then using DSP to reconstruct the original wideband content.