Part 7 of Signal Lab, a 10-part series on GopherTrunk’s offline signal-analysis workbench. Part 4 let you *see modulation quality; the VSA lets you measure it to bench-instrument precision.*
TL;DR: Signal Lab’s vector signal analyzer reports the same metrics a bench VSA does: carrier-frequency error; RMS and peak EVM, each split into magnitude and phase error; I/Q gain imbalance and quadrature skew; origin (DC) offset; an EVM-vs-symbol trace; and an error-vector spectrum. Each isolates a different defect, so instead of “the constellation looks off” you get “carrier is 1.2 kHz high and there’s 0.4 dB of gain imbalance.”
Key takeaways
- EVM is one number; the VSA is the breakdown. Splitting EVM into magnitude and phase error tells you what kind of error dominates.
- Carrier-frequency error isolates tuning from modulation quality.
- Gain imbalance and quadrature skew are front-end defects, not signal weakness — the VSA separates them cleanly.
- The EVM-vs-symbol trace finds when quality degraded across the burst.
- The error-vector spectrum shows where in frequency the error energy lives.
Cheat sheet
| VSA metric | What it isolates |
|---|---|
| Carrier-frequency error | Residual tuning offset |
| RMS EVM | Overall modulation quality (average) |
| Peak EVM | Worst-case symbol |
| Magnitude error | Amplitude component of EVM |
| Phase error | Phase component of EVM |
| I/Q gain imbalance | Amplitude mismatch between I and Q |
| Quadrature skew | Departure from 90° between I and Q |
| Origin offset | DC / carrier feedthrough |
| EVM-vs-symbol trace | Quality over the burst |
| Error-vector spectrum | Error energy vs frequency |
In this post
- What a VSA measures and why the split matters.
- Carrier-frequency error vs modulation quality.
- EVM decomposed into magnitude and phase.
- Front-end defects — gain imbalance, quadrature skew, origin offset.
- Traces — EVM-vs-symbol and the error-vector spectrum.
What a VSA measures
A vector signal analyzer treats each received symbol as a vector in the I/Q plane and measures the error vector — the difference between where the symbol landed and where it ideally should have. The length of that error vector, normalized and averaged, is EVM (error-vector magnitude). You already met EVM as a single dashboard number in Part 2; the VSA is what happens when you stop averaging everything into one figure and start asking what the error is made of.
That decomposition is the whole value. Two captures can share the same 8% RMS EVM and be broken in completely different ways — one by tuning error, one by a front-end gain mismatch, one by low SNR. A single EVM number can’t tell them apart; the VSA’s breakdown can. Signal Lab exposes the full set the way bench gear does, computed off the analyzed capture.
Carrier-frequency error vs modulation quality
The first thing the VSA separates out is carrier-frequency error — the
residual offset between where the signal actually sits and 0 Hz baseband. This is
a tuning problem, not a modulation problem, and conflating the two is a
classic mistake. A perfectly modulated signal recorded slightly off-center will
show a rotating constellation and inflated EVM even though the transmitter is
flawless; the fix is to tune it out (-auto-tune, or the receiver’s AFC), not to
chase a nonexistent modulation defect.
By reporting carrier-frequency error as its own number, the VSA lets you subtract the tuning question entirely: correct the offset, and whatever EVM remains is genuinely modulation quality. Reese’s first VSA question is always the same — is the carrier where it should be? — because until it is, every other metric is contaminated.
EVM decomposed: magnitude and phase
Once tuning is accounted for, the VSA splits EVM into two orthogonal components:
- Magnitude error — the amplitude part of the error vector: symbols landing too close to or too far from the origin along their ideal direction. It points at amplitude problems — compression, AGC misbehavior, fading.
- Phase error — the angular part: symbols rotated off their ideal phase. It points at phase noise, residual frequency error, and phase-tracking trouble.
It also reports RMS EVM (the average, your headline quality number) alongside peak EVM (the single worst symbol). The gap between them is diagnostic: RMS and peak close together means uniformly noisy; a low RMS with a high peak means mostly clean with occasional bad symbols — often a transient interferer rather than a steady-state weakness. Whether magnitude or phase error dominates tells you which physical mechanism to suspect, which is exactly the kind of pointer a lone EVM number can never give.
Front-end defects: imbalance, skew, offset
Three VSA metrics describe the receiver, not the signal — defects introduced by the I/Q front-end itself:
- I/Q gain imbalance — the amplitude of the I channel doesn’t match the Q channel. On a constellation the rails or clusters stretch along one axis; the image-rejection number from Part 2 drops.
- Quadrature skew — I and Q aren’t exactly 90° apart. The constellation shears, as if the plane were pushed out of square.
- Origin offset — residual DC / carrier feedthrough that pulls the whole constellation off center; it shows up as a spike at 0 Hz in the PSD.
These are the fingerprints of a particular SDR and gain setting rather than of the signal in the air, which is why isolating them matters: a capture with fine SNR but visible gain imbalance and skew was recorded on a front-end that needs I/Q correction, and no amount of resignal-hunting will improve it. Synthesis (Part 6) lets you inject each of these deliberately and confirm the VSA reads them back — the cleanest way to build intuition for what each defect looks like.
Traces: EVM-vs-symbol and the error-vector spectrum
The last two outputs are traces rather than single numbers, and they answer when and where.
The EVM-vs-symbol trace plots error magnitude across the burst, symbol by symbol. A flat trace means steady quality; a ramp or a spike means quality changed during the capture — a fade, a collision, a transmitter settling. This is how you catch a capture that averages “fine” but was briefly awful at one instant.
The error-vector spectrum takes the error signal itself and shows its energy versus frequency. A flat error spectrum is consistent with white noise (just low SNR); a peaked error spectrum means a specific spur or tone is injecting error at a particular frequency — an interferer or a front-end artifact you can then go find. Where the EVM-vs-symbol trace localizes error in time, the error-vector spectrum localizes it in frequency, and together they turn a raised EVM into an actionable lead.
Ada’s takeaway from her first VSA session: the dashboard’s single EVM told her something was wrong; the VSA told her the carrier was 1.2 kHz high and the rest was a clean 7% — so the fix was a retune, not a new antenna.
A worked diagnosis
Put the whole panel together on one capture and watch it name a fault. Ada’s recording reads 9.5% RMS EVM — not great, not terrible — and she wants to know what’s costing her. The VSA breaks it down:
| VSA metric | Reading | What it says |
|---|---|---|
| Carrier-frequency error | +180 Hz | Small; not the culprit |
| Magnitude error | 3.1% | Modest amplitude spread |
| Phase error | 8.8% | Dominant — the error is mostly angular |
| I/Q gain imbalance | 0.15 dB | Front-end is fine |
| Quadrature skew | 0.3° | Front-end is fine |
| Origin offset | −42 dB | Negligible DC |
The story is unambiguous: the carrier is nearly centered, the front-end is clean, and magnitude error is small — but phase error dominates. That points at phase noise or a phase-tracking loop that’s working too hard, not at a weak signal or a misbehaving SDR. Ada wouldn’t have reached that conclusion from a 9.5% EVM number or even from staring at the constellation, where a phase-heavy error just looks like a general smear. The decomposition did what the picture couldn’t: it named the mechanism. And because synthesis (Part 6) can inject phase noise deliberately, she can confirm the theory — dial in known phase noise, watch phase error rise while magnitude error stays flat, and match the signature.
That’s the VSA’s real gift. A single EVM number is a thermometer: it tells you the patient has a fever. The VSA is the full workup — it tells you which system is sick, which is the difference between “this capture is degraded” and “retune by 180 Hz and chase the phase noise; everything else is healthy.”
Where this goes next
You can now characterize a known signal to bench precision. But Mercury still won’t say what it is. Part 8 turns to the unknown: blind signal identification, the offline signal-ID reference database that names an undecodable carrier from its symbol rate and modulation, and the wideband survey — where Ada finally gets a best guess for Mercury and hands it off. The SigLab docs carry the full VSA metric list.
FAQ
How is the VSA different from the dashboard’s EVM? The dashboard gives one EVM number. The VSA decomposes it — magnitude vs phase error, RMS vs peak — and adds carrier-frequency error, I/Q imbalance, quadrature skew, origin offset, and per-symbol/per-frequency traces, so you learn what kind of error you have, not just how much.
Why split EVM into magnitude and phase? Because they point at different causes. Magnitude error suggests amplitude problems (compression, AGC, fading); phase error suggests phase noise or residual frequency error. The split tells you which mechanism to chase.
Are gain imbalance and quadrature skew signal problems? No — they’re front-end defects, introduced by the receiver’s I/Q path. The VSA isolates them so you don’t mistake a receiver imperfection for a weak signal.
What does the EVM-vs-symbol trace catch that RMS EVM misses? Transients. A capture can average a healthy EVM while being briefly terrible at one instant (a collision or fade). The trace shows that spike; the average buries it.
Series navigation
Part 7 of 10 · ←Part 6 · Next → Part 8: Naming the Unknown