FPGA Doesn’t Repeal Physics

Why Receiver Numbers Still Matter in SDRs (and How to Use the Right Ones)

A recurring bit of “computational RF” marketing claims that once a radio’s signal chain moves into an FPGA, the old receiver numbers (especially DR3) become “legacy,” and new numbers (SFDR, ENOB, sampling rate, and friends) somehow replace them.

That’s not an engineering argument. It’s a category error. SDR and FPGA-based platforms absolutely change what’s possible in usability, filtering, feature evolution, and visibility. They do not make intermodulation, overload, phase noise, or headroom constraints disappear. Those limits simply move to different points in the chain, and you still need to measure and reason about them correctly.

The “mixers to math” slogan breaks at the first non-linear device

The core mistake in “numbers don’t apply anymore” arguments is pretending that digital processing changes the laws of linearity. It doesn’t. IMD exists wherever non-linearity exists ... and every SDR still contains analog stages.

At minimum, distortion and overload mechanisms exist in:

• RF preamps / LNAs (gain compression, IMD, blocking)
• Mixers (in superhets and many hybrid architectures)
• ADC front ends (clipping, harmonic distortion, overload behavior)
• Clocking / LO phase noise (reciprocal mixing)
• Digital pipelines (overflow/saturation if mis-scaled ... usually engineered to be negligible)
• DAC + driver chain and power amplifiers (transmit IMD)

Moving selectivity and demodulation into FPGA fabric doesn’t remove these effects. It changes where the limits show up, and which tradeoffs are practical. Physics does not care whether the math runs in copper traces, code, or FPGA LUTs.

Architecture doesn’t delete limits ... it relocates the bottleneck

Receiver performance is often “the worst limiter wins.” Depending on band conditions, the limiter might be:

• Front-end overload (gain/attenuation choices, preselection, blocker handling)
• ADC headroom and distortion (what happens near full-scale, and how cleanly it fails)
• LO/clock phase noise (reciprocal mixing close to strong signals)
• Noise floor (only when the band and environment allow it to matter)

A disciplined lab report separates failure modes instead of pretending a single number explains everything. That’s why strong-signal linearity and phase-noise behavior are treated as distinct constraints.

Stop treating different metrics as interchangeable “replacements”

The framing “old ham DR3 vs new pro SFDR/ENOB/sampling rate” is a false swap. These metrics answer different questions, often under different test boundaries. Use them as a set, not as marketing chess pieces.

DR3 still matters in FPGA SDRs because IMD is not removable after it’s created

DR3 is typically derived from a two-tone test: inject two strong signals and measure the third-order products that fall near the desired frequency. This is a system-level linearity check that includes whatever analog stages are active under the test condition ... including the ADC if it is the limiting element.

The practical reason DR3 stays relevant is simple: when intermod products land inside your passband, they become indistinguishable from real signals. DSP can’t “unmix” distortion products that are now in-band.

What DR3 is not

• Not “analog-only.”
• Not replaced by sampling rate.
• Not the same as SFDR.
• Not a phase-noise metric (that’s RMDR territory).

RMDR tells you what phase noise will do to you when the band is crowded

RMDR characterizes how LO phase noise ... or sampling-clock phase noise in direct-sampling radios ... mixes with a strong adjacent signal and raises the noise floor around it. In real operating, this often becomes the dominant limitation close to strong signals, even when sensitivity numbers look impressive.

For direct-sampling SDRs, the sampling clock is effectively the LO. Clock quality and close-in phase noise behavior matter a lot because reciprocal mixing is not “fixed by DSP” once the noise has been translated into your receive channel.

What RMDR is not

• Not improved by FFT horsepower.
• Not equivalent to DR3.
• Not something you solve with “more sampling rate.”

SFDR is useful ... but it’s not a universal receiver performance replacement

SFDR is the ratio between a desired signal and the largest spurious spectral component observed within a defined bandwidth. It is particularly useful for describing spur cleanliness and “ghost management” across wider spans.

The trap is pretending SFDR replaces DR3. SFDR is often discussed in single-tone contexts, while DR3 directly reflects two-tone IMD behavior ... which maps very well to “two big stations near my weak one.”

NPR complements the classic metrics when “many signals at once” is the stress case

Noise Power Ratio testing uses notched noise to emulate wideband, multi-signal loading and measures how much noise/distortion fills the notch. It can be an excellent stress method in environments that resemble “lots of carriers all the time.”

NPR is not a magic trump card. It complements DR3/RMDR, it doesn’t replace them.

ENOB is a converter metric ... useful, but often abused in ham arguments

ENOB is derived from SINAD and expresses the effective resolution of an ADC under specific test conditions. It can hint at converter class and implementation quality, but it does not automatically include front-end filtering strategy, gain distribution, band-to-band behavior, or real-world spur management.

ENOB is part of the story. Using it as a substitute for DR3/RMDR is mixing boundaries.

Sampling rate is not a distortion eraser ... it’s a tradeoff knob

Higher sampling rate can enable oversampling and processing gain. After decimation, in-band noise density can drop, and narrow filters can make weak signals visible with enough integration time. That’s real.

What it does not do is increase ADC full-scale headroom or prevent overload and IMD generation in the analog chain. Processing gain helps you with noise. It does not cancel nonlinearity.

What SDR and FPGA architectures really do better than classic superhets

The advantages are real ... just not the “old numbers are defunct” fantasy.

Where selectivity happens

In many SDR designs, you digitize a wider chunk first and apply steep selectivity digitally. This enables sharp, stable filters, variable bandwidths, multiple slice receivers, and workflow that classic fixed-path radios often cannot match. (The front end and ADC still decide whether strong signals crush you.)

Feature evolution over time

SDR pipelines are reconfigurable. Vendors can refine DSP, AGC behavior, notch tools, and filtering without redesigning the entire radio. That’s a genuine architectural advantage.

Spectrum visibility and operating effectiveness

Panadapters and waterfalls can reduce tuning time, improve split operation, and make openings easier to spot. This is a usability win ... not proof that strong-signal limits vanished.

Some adaptive transmit capabilities

Predistortion and linearization can improve transmit cleanliness under certain conditions. The key point: predistortion compensates for nonlinearity ... it doesn’t repeal it.

Myth vs reality ... the argument-stoppers

• Myth: DR3 is a legacy analog heuristic. Reality: DR3 measures system non-linearity under two-tone stress. SDRs still have analog front ends and ADC limits.
• Myth: SFDR replaces DR3. Reality: SFDR describes worst-spur behavior; DR3 describes two-tone IMD behavior. Related, not interchangeable.
• Myth: ENOB is receiver performance. Reality: ENOB is a converter metric under defined conditions ... not a whole-radio result.
• Myth: Higher sampling rate erases distortion. Reality: Oversampling can lower in-band noise density, but it does not prevent clipping, IMD, or reciprocal mixing.
• Myth: The waterfall proves higher linearity. Reality: Displays reflect processing gain, bin bandwidth, averaging, spur behavior, and scaling ... not overload points.

A sane buyer’s checklist that respects SDR benefits and RF physics

Know your environment

• If your local noise floor is high, tiny MDS differences won’t matter much.
• If you operate near strong adjacent signals, close-in DR3 and RMDR matter a lot.

Look for multiple strong-signal metrics

• DR3 at multiple spacings (close-in and wider)
• RMDR at similar spacings
• Blocking dynamic range (where available)
• Phase-noise plots when published

Ask where “roofing protection” really happens

• Is there robust RF preselection or band-pass filtering ahead of the ADC?
• Is gain/attenuation strategy documented (and sensible) for strong-signal conditions?

Use SFDR and NPR as complements

• SFDR ... spur cleanliness and “ghost management” across bandwidth
• NPR ... multi-signal loading behavior when many carriers are present

Treat ENOB and sampling rate as design hints, not winner labels

They can suggest converter class and processing-gain possibilities ... without pretending they cancel overload, IMD, or reciprocal mixing.

Closing: Architecture matters. Physics sets the limits. Numbers are not defunct.

FPGA-based SDRs are powerful because they move selectivity, demodulation, and features into a domain where change is cheap and performance can evolve.

But the front end, converter headroom, and LO/clock phase noise still decide what happens when the band is crowded. The correct stance isn’t “DR3 is obsolete,” or “SFDR/ENOB replace everything.”

The correct stance is: use DR3 for IMD, RMDR for reciprocal mixing, SFDR for spurs, NPR for multi-signal loading, and ENOB plus sampling rate to understand converter limits and processing-gain possibilities ... without pretending they repeal physics.

Interested in more technical content? Subscribe to our updates for deep-dive RF articles and lab notes.

Questions or experiences to share? Feel free to contact RF.Guru via our RF.Guru contact page for technical support and feedback.