Overview
This article takes a deep dive into the Roland/BOSS GK pickup and its crosstalk performance, with a focus on the GK-5. Crosstalk affects audio-to-MIDI performance and other signal processing workflows that rely on pitch detection. In audio-to-MIDI conversion, crosstalk can trigger false or ghost notes when the pitch detector misidentifies a note-on from bleed between strings.
As far as I can tell, there’s no definitive, published study on the Roland/BOSS GK pickup’s crosstalk performance. The best I’ve found are unsupported figures, typically cited as around 35–40 dB of crosstalk rejection. I haven’t been able to locate any measurements backed by a clearly documented, publicly available study.
Let’s do some real-world testing.
Test Procedure
I’m testing the GK-5 Internal Kit installed in our Iron Man test guitar. The GK-5 pickup height is set to an optimal position: about 1 mm from the strings and around 20 mm on average from the bridge (accounting for intonation), just beside the Floyd Rose floating tremolo. Per-string pickup sensitivities were calibrated in Apple Logic Pro, with the BOSS GM-800 acting as the audio interface. Logic Pro’s level meters provide accurate dB metering that far exceeds the meters on the GM-800 itself. Using proper gain staging, I adjusted the GM-800’s per-string level controls to land at a sweet spot of no more than -1.5 dB on hard string hits.
Here are the optimal levels for our setup, per the GM-800’s GK Settings:
| STR1 | 40 |
| STR2 | 35 |
| STR3 | 30 |
| STR4 | 40 |
| STR5 | 30 |
| STR6 | 30 |
All six strings were then recorded simultaneously while I picked each string in turn, carefully muting the others to prevent sympathetic resonance and mechanical crosstalk through the nut and bridge. We care most about note onsets, so each string was manually damped after about a second.
The resulting raw six-channel recording was saved as a single 6-channel WAV file for analysis in Adobe Audition. We then extracted the minimum, maximum, average, and median crosstalk for each string, relative to every other string.
Here is the audio file so anyone can replicate and verify my analysis: All-Strings.wav
Captured Signals
Here’s the screenshot of the recording:
Crosstalk appears as small perturbations in the neighboring channels, while each string’s intended signal remains clearly the most prominent. Click the image to expand.
Let’s zoom in on the crosstalk so we can take a closer look at what’s really going on. Click the image to expand.
Measurement Results
True Peaks (dBTP)
| Excited ↓ / Measured → | Low E | A | D | G | B | High E |
|---|---|---|---|---|---|---|
| Low E | -13.55 | -36.95 | -59.69 | -61.45 | -64.94 | -63.89 |
| A | -58.34 | -13.29 | -51.79 | -68.75 | -69.99 | -70.55 |
| D | -54.56 | -55.07 | -10.79 | -38.78 | -66.08 | -63.75 |
| G | -57.12 | -52.46 | -46.49 | -14.43 | -42.33 | -66.66 |
| B | -57.17 | -61.04 | -58.62 | -49.47 | -9.88 | -55.90 |
| High E | -53.48 | -54.26 | -56.25 | -62.86 | -49.08 | -13.48 |
The numbers in red show the true-peak levels for each excited string. The other values on the same row show the corresponding peak levels measured on the remaining strings, indicative of crosstalk.
Crosstalk Statistics (dB)
| String | Worst (Max) | Best (Min) | Average | Median |
|---|---|---|---|---|
| Low E | -23.40 | -51.39 | -43.83 | -47.90 |
| A | -38.50 | -57.26 | -50.59 | -55.46 |
| D | -27.99 | -55.29 | -44.86 | -44.28 |
| G | -27.90 | -52.23 | -38.58 | -38.03 |
| B | -39.59 | -51.16 | -46.56 | -47.29 |
| High E | -35.60 | -49.38 | -41.71 | -40.78 |
The results are revealing. The minimum crosstalk we observed is very low, ranging from -49 dB to -57 dB On average, it falls between -42 dB and -51 dB. Those numbers are admirable. Unfortunately, while they look good on paper, they’re not the numbers that matter most in practice.
The real issue is the peaks: brief, high-level bursts of crosstalk that typically appear near the onsets of adjacent strings. They may be short-lived, but they can be strong enough to fool the pitch and onset detector at note-on, leading to mis-triggers, ghost notes, and onsets assigned to the wrong string or even the wrong pitch. Unfortunately, these peaks bias the results in the worst possible direction.
In the end, peak crosstalk is what really matters. Based on our tests, and looking at peak crosstalk levels across strings, we measured a worst case of about -23 dB and a best case of about -40 dB for the GK-5 pickup.
At first glance, a −23 dB figure may seem concerning, but it’s important to remember what that number represents. In the context of thresholding (or gating), the onset detector must decide, using heuristics, whether an event represents a legitimate note-on or simply crosstalk. One such heuristic relies on a simple observation: when a performer excites two or more strings at the same time, their levels usually fall within the same ballpark. It is very difficult to strike one string very hard while another is extremely soft at that exact instant.
Given this, a smart onset detector considers the relative, not absolute, levels of all potential onsets and rejects candidates that fall sufficiently below the dominant excitation. A relative margin of about 12 dB from the loudest excitation is a reasonable target. For an excitation of −13.6 dBTP, as in the test results above, this places the dynamic crosstalk rejection threshold at −25.6 dBTP—comfortably above the highest measured crosstalk level of −36.95 dBTP.
Note: 12 dB corresponds to ~4× amplitude difference.
If there’s one thing I can recommend to anyone using the GM-800, or any pitch-to-MIDI device for that matter, it’s to optimize your gain staging. Place the pickup as close to the strings as practical, ideally no more than 1 mm away. And don’t simply accept the default level settings. Use a reasonably accurate meter, as we did with a DAW’s level meter (highly recommended because you’re working with true dBFS rather than arbitrary units). Take the time to dial in the per-string levels properly. You’ll likely be pleased with the results.
But can we do better?
Impressive analysis !!!! Food for thought !!! I guess that this was done on an electric guitar, where the crosstalk is the electromagnetic signal affecting the adyacent mics. I wonder how this works for nylon strings..RMC/Graphtech hex piezo pickups. The crosstalk then must be in the form of vibrations from one piezo that makes it into the other piezopickups..maybe through vibration in the bridge. Is it a better or worse situation for pitch detection? In any case, great article !! Thanks for sharing your knowledge with us !
Piezo is of no interest to me, for obvious reasons. I do know that piezo pickups have the worst crosstalk.
Very interesting, great advice. Did you just do one measurement or multiples? I’m curious how much variance you may have seen in your repeated measurements.
Yes, I did a couple of takes. There’s quite a bit of variance, but not significant in the db domain, and the tests are quite repeatable.
I’m sure that the fact that people play resonantly also affects crosstalk adversely.
Well, there are various ways to mitigate resonant mechanical crosstalk too, It has its own characteristic slow attack swell. I can show some screenshots if needed.
But a smart system can probably just use that information to aide in the detection. It’s resonant and musical anyway.
Realize the GK5 divided pickup is ideal at 20mm distance away from bridge, any closer results in less signal output due to smaller vibrating string excursions closer bridge
I understand that, Steve. And actually, the distance is, on average, 19.9 mm, accounting for intonation. I should correct my post. I just measured the closest distance.
Here are the actual numbers:
I did some testing on distances from 20 to 30 mm, and based on those results, the difference isn’t significant. The signal does get stronger farther from the bridge, but the crosstalk increases as well. The result, in terms of dB, stays more or less the same.
I’ll post this to the VG forum so we can discuss it there.