• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Comparing aliasing in three ADCs.

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
21,466
Likes
39,133
Thought it would be interesting to see how much aliasing there is in some ADCs. The three ADCs compared were a Focusrite 18i20 recording interface, Focusrite Forte recording interface, and Antelope Audio Zen Tour recording interface. The source of the signals used is a TC IMpact Twin running at 192 khz.

The first signal is -4db white noise covering 0-96khz. I used a brickwall high pass filter starting at 27 khz, -96db at 24 khz and -120 db at 23 khz. I wanted to see what would show up in these three ADCs running at 48 khz sample rates and covering 24 khz of sound. Below are spectrograms of that for each ADC. The spectrogram uses 2k FFT and goes to a light gray background at a signal level of -120 dbFS. Anything below -120 db will not show up. The 4th spectrogram at the bottom is the signal being recorded.

white noise aliasing.png


The top two show an elevated noise floor in the presence of the ultrasonic noise. Along with stronger aliasing in the upper two khz of the band. The noise floor is raised about 3 to 4 db which isn't a lot. The noise floor ignoring the highest frequencies is still several decibels below -100 db. More interesting is the Zen Tour which has its noise floor raised about 8 db on average, but also that noise floor is undulating with frequency by about 6 db between low and high points. The Zen Tour is very good with strong signals in the 24 khz band not modulating the noise floor at all. Yet ultrasonics apparently can via aliasing. Even then it fortunately has a low noise floor and this elevated level is still a few db below -100 dbFS at worst.

Next I used a -1dbFS single tone sweep from 2 hz to 96 khz while the ADCs recorded at a 48 khz sample rate.
single tone sweep aliasing.png


Here you see the distortion harmonics of the main signal until you reach 24 khz. You see the stronger aliasing at the top of the 24 khz band, and then aliasing folding over every 24 khz above the 24 khz band of the ADC. The Zen Tour shows the highest levels of the aliasing products reaching -88db FS in a few places. Though mostly it skirts around -100 dbFS. The Forte in the second spectrogram shows some additional aliasing. It is centered around 31.6 khz. This device has the same effect even if used sampling at 96 khz or 192 khz. It will act like there is aliasing around 31.6 khz which is rather strange behavior. It mostly also meanders near the -100 dbFS level.

Next I used a twin tone sweep with tones 1 khz apart for the test signal.

twin tone IMD aliasing.png



The main thing worth noting is even though the signal is ultrasonic it still shows intermodulation difference products at the 1 khz difference down in the audible band. I believe this is in the analog stage which has 100 khz of bandwidth in these devices and the intermodulation is showing up which gets recorded by the ADC. In a few places this IMD difference signal reaches - 88 dbFS though mostly it lies in the mid 90s to -100 dbFS.

Now one final spectrogram where I extended the range to -144 dbFS and used a 32k FFT.
140 db narrow spectrogram.png

You don't see anything new show up for the Zen Tour or Forte ADC, but there some curious things in the 18i20 ADC. Firstly there are low level constant tones spaced exactly 1 khz apart. These are around -126 to 128 dbFS so not really a problem, but strange nonetheless. These are present with signal or silence at all times.

Even more odd is in the presence of a strong signal in the audible band, but not with ultrasonics are tones from 125 hz and spaced 125 hz apart up to the point of 625 hz. These tones come and go related to the strength of the strongest signal. All are the same strength. At worst they reach - 106 dbFS so not a big concern, but again strange how that must be happening.

All of these ADCs are good about not modulating the noise floor or having too many artifacts when there is no ultrasonic content. However, we do see them perturbed a little when ultrasonics are present. How important is this? Many condenser microphones will have some considerable response into the mid 30 khz range. Some instruments do put out considerable levels at these frequencies especially important if they are close miked. There are microphones that have extended bandwidth like Earthworks for one example. If you were doing a line level recording of an SACD or DSD the ultrasonic rise in noise could show up as slightly elevated noise at lower frequencies in these ADCs. Sometimes during recording lighting or switching supplies or other devices can leak high levels of ultrasonics into the microphone path. That might show up at low levels. Of course one could record at 96 khz reducing all of these aliasing related issues quite a lot nearly to the point of being a complete non-issue. My test signals were something of a worst case scenario and the resulting audible band spurious signals are never high enough in level to be audible unless heard at very high gain level against a silent background. Even then not likely to be heard.
 
Nice and interesting work, thank you!
 
All of these ADCs are good about not modulating the noise floor or having too many artifacts when there is no ultrasonic content. However, we do see them perturbed a little when ultrasonics are present. How important is this? Many condenser microphones will have some considerable response into the mid 30 khz range. Some instruments do put out considerable levels at these frequencies especially important if they are close miked. There are microphones that have extended bandwidth like Earthworks for one example. If you were doing a line level recording of an SACD or DSD the ultrasonic rise in noise could show up as slightly elevated noise at lower frequencies in these ADCs. Sometimes during recording lighting or switching supplies or other devices can leak high levels of ultrasonics into the microphone path. That might show up at low levels. Of course one could record at 96 khz reducing all of these aliasing related issues quite a lot nearly to the point of being a complete non-issue. My test signals were something of a worst case scenario and the resulting audible band spurious signals are never high enough in level to be audible unless heard at very high gain level against a silent background. Even then not likely to be heard.

Utterly fascinating. When I worked briefly making recordings of classical music performances many years ago now, I noticed that chamber recordings made with the Earthworks QTC1s (now the QTC30s, I think) sounded distinctly better at 96 kHz with the Motu ADC that I was using. Perhaps you have found the cause of this admittedly subjective observation.
 
@soundArgument post convinced me to check out 96 khz aliasing since it was fairly easy to do.

First white noise high pass filtered at 27 khz.

white noise aliasing 96.png

Again the presence of noise above the band covered by the ADC raised the noise floor some. The two Focusrite ADCs were up about 3 db just like when working at 48 khz sample rates. Of course in actual use with there will be no raised noise floor until you get above 48 khz instead of 24 khz in the previous test.

The Zen Tour still has a rippled slightly raised noise floor. However, instead of being raised about 8 db, it was only 3 db here. The variation with frequency is reduced to only about 2 db. So some improvement there.

Here is a single tone sweep.
single tone sweep aliasing 96.png

I'm not showing it again, but the 18i20 still has idle tones spaced at 1 khz intervals at about -126 dbFS. Only now they extend to 48 khz. The 18i20 also has the 125 hz to 625 hz spaced tones in the presence of the sweep which disappear when the tone exceeds the nyquist band of the sample rate. If you look very closely you can see the noise floor bumps up 3 db extra once the sweep exceeds nyquist in addition to the aliasing tones you see.

At 96 khz sample rate the Forte shows the idle tone it has at 31.6 khz. I noticed that archimago's Forte has the same idle tone only his is at 37 khz. You also can see all signals alias around that 31.6 khz point which is odd behavior. These alias signals are mostly at or below -100 dbFS. You also have the aliasing from exceeding the 96 khz sample rate which is higher in level. It varies about -90 to -98 dbFS.

The Zen Tour has about the same aliasing above nyquist varying around the mid -90 dbFS range. There is the barely visible 16 khz idle tone when connected to my Macbook. I was using a thunderbolt connection so it isn't USB related. Also it is actually 16.4 khz. No aliasing or other spurious signals related to that. Not much to comment upon otherwise.

Unless using microphones like Earthworks with extended frequency response the majority of condenser microphones have the frequency response die off rapidly somewhere around the mid-30 khz range. So by using 96 khz sample rates which give 40 khz response, you'll not experience aliasing at all with most gear. The exception being switching power supplies or dirty lighting systems radiating something into the device.
 
This is very interesting, since I was about to open a thread asking "how does A/D performance compare to DACs these days?"
- Can you explain the top horizontal scales? I'm not grasping those and the varied numbers across the top.
- How is the raw noise floor compared to DACs? (If you can do that in a format like @amirm for comparison).
- These DACs are multibit or 1-bit or hybrid? What is inside?
 
This is very interesting, since I was about to open a thread asking "how does A/D performance compare to DACs these days?"
- Can you explain the top horizontal scales? I'm not grasping those and the varied numbers across the top.
- How is the raw noise floor compared to DACs? (If you can do that in a format like @amirm for comparison).
- These DACs are multibit or 1-bit or hybrid? What is inside?
I'll answer this as best I can a little later.
 
The numbers at the top are just seconds. I played a 60 second sweep tone.

These DACs and ADCs were all sigma-delta units. Oversampling with a few bits.

In general, good ADC's are around 10 db more noisy than the very best DACs so Sinad numbers are worse about that much. THD can be just as low as DACs though it is not always.

Does that make sense or do you have other questions?
 
- These DACs are multibit or 1-bit or hybrid? What is inside?
Did you mean to write ADCs here?

If memory serves:
Scarlett 18i20 is a CS4272 grave, like all Scarletts. (@Blumlein 88 - this was the 1st gen, I guess? I won't be surprised if later models showed some improvements when it comes to the low-level interference. Tones spaced 1 kHz smell of USB, for example.)
Forte: AK5388 (ADC) + CS4398 (DAC).
Zen Tour: No clue. Apparently there were two revisions, converter S/N 120 dB before 2017 and 124 dB after. (Maybe AK5388 before and AK5578 with 2 channels in parallel after?)

Filter stopband rejection spec is -80 dB for the AK5388 vs. -85 dB for the AK557x family and -92 to -97 dB for the CS4272 ADC, so if the Zen Tour does in fact use an AK5578 that would check out.
 
Just when I thought that digital audio was perfect. :p
 
Yes 1st Gen Scarlett and Zen tour.
 
Back
Top Bottom