Author |
Message |
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Sun May 08, 2005 7:20 pm Post subject:
Digital synth aliasing comparison |
|
|
I used OscilloMeter for these tests. A Blackmann/Harris 4 window and 'flat' weighting were used. The input device is an EMU 1820m.
My initial motivation was to see if a square is indeed a square at 12.5 kHz for a variety of digital synths and to see how much aliasing occurs. I was surprised, to say the least.
pics removed due to their inaccuracy Last edited by cebec on Wed May 11, 2005 7:12 am; edited 3 times in total |
|
Back to top
|
|
|
ian-s
Joined: Apr 01, 2004 Posts: 2669 Location: Auckland, New Zealand
Audio files: 42
G2 patch files: 626
|
Posted: Sun May 08, 2005 11:10 pm Post subject:
|
|
|
That looks like quite a bit of aliasing on the analog square
The G2 is almost a sine but the next harmonic of a 12K5 square is 50K, well past nyquist so it should look like a sine if it has been band limited?
Thanks for the data. |
|
Back to top
|
|
|
Kassen
Janitor
Joined: Jul 06, 2004 Posts: 7678 Location: The Hague, NL
G2 patch files: 3
|
Posted: Mon May 09, 2005 6:40 am Post subject:
|
|
|
Do I understand correctly that you used the soundcard to record the software synths too? Something seems wrong here; Tassman should not alias, especially not on a simple quare; A.A.S. spend large amounts of time and care combating aliassing. _________________ Kassen |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Mon May 09, 2005 7:30 am Post subject:
|
|
|
Yes, I used the EMU 1820m with the following specs:
The op amps are JRC 3292B. Two in the two output channels section and four in the input channels section. The input channels section have two relays for switching between +4dB and -10dB. The ADCs are the AKM AK5394AVS and DACs Cirrus Logic CS-4398-CZ.
I set up the PatchMix DSP applet to route the audio out of the softsynth (ASIO Out 1/2) to Wave In 1/2 (Oscillometer). The only time conversion took place was when the G2 and the analog square were measured.
I'm not trying to start another 10 page thread but I was curious and thought I'd share what I found. I'm still learning... |
|
Back to top
|
|
|
Kassen
Janitor
Joined: Jul 06, 2004 Posts: 7678 Location: The Hague, NL
G2 patch files: 3
|
Posted: Mon May 09, 2005 10:34 am Post subject:
|
|
|
Yes, that should work alright. _________________ Kassen |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Mon May 09, 2005 9:06 pm Post subject:
|
|
|
I've done some further tests. A 12.5 kHz rectangle is not as ideal as a narrow pulse. And, at 12 kHz, a rectangle does not generate nearly as much aliasing in any of the synths examined as does 12.5 kHz!
An impulse, narrow square (10%) or falling sawtooth at 1, 6, 12, and 12.5 kHz seem much more interesting in terms of the aliasing they can produce. The Nord G2 has a particular problem with these types of waveforms. The aliasing generated reaches further into the audible range than any other digital synth I tested. I'm not sure if this is the antialiasing algorithms or the G2 DACs.
I'm afraid I'm not too hopeful since an identical test using the G2 Demo software did not result in any of the same aliasing. I tested the demo software in internal loopback and in loopback with an analog cable.
The aliasing pictured in the G2 hardware graph increased in intensity from these waveforms at 1.3 kHz and up.
pics removed due to their inaccuracy Last edited by cebec on Wed May 11, 2005 7:26 am; edited 1 time in total |
|
Back to top
|
|
|
ian-s
Joined: Apr 01, 2004 Posts: 2669 Location: Auckland, New Zealand
Audio files: 42
G2 patch files: 626
|
Posted: Tue May 10, 2005 12:18 am Post subject:
|
|
|
That’s very strange, 12K is exactly 1/8th of 96K so it should not alias at all. I repeated the test and discovered two things.
1. I show no aliasing at exactly 12KHz 10% rectangle sampling at 48K but plenty if I vary a little from the 12K frequency.
2. I can only just hear a 12KHz tone, though my wife complained from the other side of the apartment.
I’m using a $60 USB audio device and the freeware VisualAnalyser VA.EXE |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Tue May 10, 2005 6:56 am Post subject:
|
|
|
Thanks for the tip on VisualAnalyzer! I hadn't come across this in my search, for some reason. I'll try the same tests with this, later. |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Tue May 10, 2005 10:01 am Post subject:
|
|
|
Weird. I'm doing some tests with the G2 Demo on a 48 kHz motherboard sound chip and VisualAnalyzer 6. The aliasing isn't present at 12 kHz, but when I sweep downward, it becomes more pronounced. This is with the same waveforms as above.
However, lowering the amplitude of the signal causes the aliasing to decrease significantly...Moreso than I'd expect in proportion to the amount of attenuation. Could I be seeing limits of these two analyzers with these tests, somehow? Or reading them incorrectly?
VisualAnalyzer has an input meter whereas OscilloMeter has a pair of overload indicators. It seems if the input of VisualAnalyzer is clipping or close to maximum, the 'aliasing' is present. Back off a bit and it's gone...
Update: Tried it with MAX/MSP and Bidule, too... Same thing. It seems I've been 'clipping' the analyzer input on OscilloMeter but its overload indicators weren't lighting because the peak meters say the signal's at -14 db. Will finish these infernal tests tonight when I get back to my other rig. |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Tue May 10, 2005 3:09 pm Post subject:
|
|
|
Ok. Pretty much the end of the tests.
Measured G2 hardware with VisualAnalyser at 96 kHz SR with narrow pulses at a wide range of frequencies.
To sum up, there's something really wrong with the G2s narrow pulses. No other synth I've tested shows the amount of aliasing I can see here. Whether or not we can hear this and whether it's present in other signals moreso on the G2 than other digital synths is up for debate. It seems like the 'problem' can be remedied if it's a matter of tweaking the oscillator code.
Thanks. |
|
Back to top
|
|
|
jksuperstar
Joined: Aug 20, 2004 Posts: 2503 Location: Denver
Audio files: 1
G2 patch files: 18
|
Posted: Tue May 10, 2005 5:16 pm Post subject:
|
|
|
I've been hard pressed to get a good punchy and bubbly sound out of the G2, without giving too much attack and getting too close to "glitch" sounds. Alot of sounds I'm happy with, but it tends to be a "soundblaster" like sound, instead of a ript off your pants and wax your behind type of sound.
I'm almost about to agree with Kassen on this.
I would assume that Demo code vs. the actual G2 code is based on the very same source. Which if this is true, most aliasing we hear is coming from the G2's DACs. Maybe there's more testing on my part to be done using the DACs of my Layla24 audio card (or the DACs of a Lexicon MPX-1). I'll put what I find here. someday. |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Wed May 11, 2005 5:36 am Post subject:
|
|
|
I removed the pictures because my original tests were flawed for two reasons: clipping the analyzer input and feeding the analyzer frequencies that did not match what I was reading on the G2s oscillator modules. Once I synced the two, using VA's frequency counter, I got results more in line with expectations.
Of course, there are so many variables involved with the DACs and my audio card's ADCs it's hard to separate 'signal' from 'noise'
My conclusion, then, is that the aliasing with high frequency narrow pulse and similar waveforms is confined to the G2 hardware.
Multiple tests with the G2 demo software and the hardware have provided the same repeatable results Anyone can perform this test with a 25% or 10% pulse measured at any sampling rate with either OscilloMeter or VisualAnalyser (both free downloads).
I would like to learn more about the G2s DACs, though, and if any DSP code changes could compensate for 'flaws' in the hardware layer. Does anyone know the model of the DACs? Last edited by cebec on Wed May 11, 2005 3:15 pm; edited 1 time in total |
|
Back to top
|
|
|
paul e.
Joined: Sep 22, 2003 Posts: 1567 Location: toronto, canada
Audio files: 2
|
Posted: Wed May 11, 2005 6:54 am Post subject:
|
|
|
wouldn't you rather just make music ? _________________ Spiral Recordings |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Wed May 11, 2005 7:03 am Post subject:
|
|
|
Yes, of course, but I rather like doing a lot of different things. |
|
Back to top
|
|
|
mosc
Site Admin
Joined: Jan 31, 2003 Posts: 18197 Location: Durham, NC
Audio files: 212
G2 patch files: 60
|
Posted: Wed May 11, 2005 7:57 am Post subject:
|
|
|
Cebec, thanks for removing the pics. I personally applaud your persistent curiosity about quality of what comes out of digital instruments. Even though some of these measurments were taken in error, there was still a lot to be learned.
Respect... _________________ --Howard
my music and other stuff |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Sun Jun 26, 2005 11:51 am Post subject:
|
|
|
Some large spikes at 20 kHz I observed on the G2 hardware seemed to be the D/A anti-aliasing filter --
Later, when comparing the same patch on the G2 Demo (with E-MU 1820m) I observed the same phenomenon.
Therefore, the phenomenon seems to reside in the code rather than the G2 DACs, specifically. |
|
Back to top
|
|
|
Afro88
Joined: Jun 20, 2004 Posts: 701 Location: Brisbane, Australia
Audio files: 12
G2 patch files: 79
|
Posted: Sun Jun 26, 2005 3:12 pm Post subject:
|
|
|
|
|
Back to top
|
|
|
Kassen
Janitor
Joined: Jul 06, 2004 Posts: 7678 Location: The Hague, NL
G2 patch files: 3
|
Posted: Sun Jun 26, 2005 3:28 pm Post subject:
|
|
|
jksuperstar wrote: |
I'm almost about to agree with Kassen on this.
|
Heresy!
;¬)
I think it´s great the Cebec is doing this. These problems are realy the secret underbully of the digital synth industry and a large factor in the sound. It´s not just the G2; many modern synths suffer from this and I think it´s time manifacturers tried a little harder to fix this instead of concentrating on millions of voices and gigs of samples.
To be honest; I´m kinda happy it´s looking to be in the code. This means it would be fixable in theory (but it would take a complete redisign). I think a G2 could make me very happy if this would be fixed. For sure there are other problems but some clever patching and some thought could get around those most of the time. With less aliassing the room would also be cleared for more complex patches that would still be stable. _________________ Kassen |
|
Back to top
|
|
|
Kassen
Janitor
Joined: Jul 06, 2004 Posts: 7678 Location: The Hague, NL
G2 patch files: 3
|
Posted: Sun Jun 26, 2005 3:30 pm Post subject:
|
|
|
I forgot to praise Cebec for doing independant research. Of cource it´s a bit rough around the edges, but it´s honest and it´s certainly being peer-reviewed. Science isn´t just for universities and large comercial companies. _________________ Kassen |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Sun Jun 26, 2005 6:05 pm Post subject:
|
|
|
Thanks, Kassen!!
I've been interested, lately, in occasionally being concerned with 'patching around' the limitations or deficiencies in these systems... aliasing, for example. but moreover, i'm learning and having fun, more or less.
In the end, though, I, too, wish the manufacturers would place more stock in manufacturing quality, documentation, and customer support than in 'competitive' features (voice count, googleplex-saws, etc.), especially in a niche market such as this. Ah, idealism... |
|
Back to top
|
|
|
Kassen
Janitor
Joined: Jul 06, 2004 Posts: 7678 Location: The Hague, NL
G2 patch files: 3
|
Posted: Sun Jun 26, 2005 6:59 pm Post subject:
|
|
|
Yeah, Paul´s point is tempting but I think that for electronic music learning about systems can be very comparable to practicing on a acoustical instrument and of cource dealing with questions can be very inspirational.
Every system has limitations, charting them out will only help you play it better. Aliassing can be very musical, i think it´s a large element in the energy of some micro music, but it realy helps to know how it works and how to work with it. If you know the sample frequency you could play in some home-brew "key" that makes the sample frequency "in tune", for example.
And, well, isn´t it fun to laugh at keyboards aimed loudly at keyboardists that have more voices then they have keys? I think that´s about as funny as people who buy a three gigaHz machine to be able to "internet more quickly". For my cash I´d rather have stuff I actually need.
I think there is most certainly room for idealism; many of these systems are labours of love more then pure commercial products. Compared to the leads and the Electro Clavia certainly isn´t getting rich off the Modulars, Millar Puckete is giving away P.D. and I happen to know A.A.S. is very much concerned about questions of how to controll synths that go quite a bit beyond what the market is currently asking for. Sollutions are a little harder to come by; it´s one thing to find a VCA can introduce aliassing, it´s quite another to prevent it, even if your OSC´s are "safe". Suppose also we both make systems, yours won´t have aliassing and mine has. The fastest attack of your envelopes will be slower then mine (given the same sample rate) and I may proudly advertise this as a advantage.....
Once again we are left to nod our heads in contemplation and simply mutter that it´s hard. :¬) _________________ Kassen |
|
Back to top
|
|
|
cebec
Joined: Apr 19, 2004 Posts: 1098 Location: Virginia
Audio files: 3
G2 patch files: 31
|
Posted: Mon Jun 27, 2005 5:09 pm Post subject:
|
|
|
jksuperstar, somewhere in the nm forum, suggested lowpass filtering control signals in order to reduce aliasing.
recent tests have given good results -- a ~20 dB attenuation of the ~24 kHz 'spikes', using a level multiplier as a 'VCA' and lowpassing the amplitude envelope with a 6 dB slope and cutoff at 1.05 kHz in a bass drum patch, for example. |
|
Back to top
|
|
|
|