Hi
Sinewave spectral measurements on three different RME Multiface 1 showed some expected static spikes emerging from the noise carpet as harmonic distortion, along with some others, unexpected static spikes not correlated to the input signal. And also, there were some rather strange, shifting spikes, steadily moving upwards on the frequency scale while the device was warming up.
For further investigation and better delta temp during the first minutes of poweron, I therefore cooled the Multiface box down to 10°C before connecting it to the power supply for ne next set of measurements. Note that the graphs show cumulative max values measurements over 30 (early measurement over the first 10 minutes) and over 100 cycles (up to a total monitoring time of approx. 1 hour). The measurement were done with a sampling rate of 88k2. Switching to a sampling rate of 96k showed a slightly different spikes pattern.
MF1_Dynamic.gif
On the gif you will see the main spike of the 1.5kHz looped-back test signal along 3 spikes of it's 2nd, 3rd and 5th harmonic distortion.
Then there are these other spikes which are not related to the test signal, but certainly related to the DA-AD process (Looped back SPDIF instead does not show any spikes except of the test signal): Such an unexpected, static cluster of spikes is seen at > 20kHz. And then there is the wild bunch of these shifting spikes. The speed of theirs migration is higher during initial warmup, steadily slowing down while the device gets warm. Definitely very strange seems the spike emerging at approx. 250Hz, but only after some 10 minutes of warmup and then steadily wandering upwards the frequency scale, reaching some 1200Hz after an hour or so.
Does anybody have an idea about these thermally dependent, shifting artefacts? I would like to understand the cause of these artefacts in order to eventually get rid of the worst of them (as is, dominating the "regular" THD products of the DA-AD process).
Sinewave spectral measurements on three different RME Multiface 1 showed some expected static spikes emerging from the noise carpet as harmonic distortion, along with some others, unexpected static spikes not correlated to the input signal. And also, there were some rather strange, shifting spikes, steadily moving upwards on the frequency scale while the device was warming up.
For further investigation and better delta temp during the first minutes of poweron, I therefore cooled the Multiface box down to 10°C before connecting it to the power supply for ne next set of measurements. Note that the graphs show cumulative max values measurements over 30 (early measurement over the first 10 minutes) and over 100 cycles (up to a total monitoring time of approx. 1 hour). The measurement were done with a sampling rate of 88k2. Switching to a sampling rate of 96k showed a slightly different spikes pattern.
MF1_Dynamic.gif
On the gif you will see the main spike of the 1.5kHz looped-back test signal along 3 spikes of it's 2nd, 3rd and 5th harmonic distortion.
Then there are these other spikes which are not related to the test signal, but certainly related to the DA-AD process (Looped back SPDIF instead does not show any spikes except of the test signal): Such an unexpected, static cluster of spikes is seen at > 20kHz. And then there is the wild bunch of these shifting spikes. The speed of theirs migration is higher during initial warmup, steadily slowing down while the device gets warm. Definitely very strange seems the spike emerging at approx. 250Hz, but only after some 10 minutes of warmup and then steadily wandering upwards the frequency scale, reaching some 1200Hz after an hour or so.
Does anybody have an idea about these thermally dependent, shifting artefacts? I would like to understand the cause of these artefacts in order to eventually get rid of the worst of them (as is, dominating the "regular" THD products of the DA-AD process).
Attachments
Last edited:
Sigma-delta idle tones or clock harmonics being downconverted to audio frequencies?
You wrote about a DA-AD process. If the DAC and ADC have independent clocks, you might get a mixing/aliasing product at the n-th harmonic of the DAC clock minus the m-th harmonic of the ADC clock. As the clocks are independent, the difference between those harmonics might drift over time, especially when things are warming up.
Besides, sigma-delta ADCs may have idle tones around odd multiples of fsample/2 that depend on offsets, and those offsets may also drift with temperature.
You wrote about a DA-AD process. If the DAC and ADC have independent clocks, you might get a mixing/aliasing product at the n-th harmonic of the DAC clock minus the m-th harmonic of the ADC clock. As the clocks are independent, the difference between those harmonics might drift over time, especially when things are warming up.
Besides, sigma-delta ADCs may have idle tones around odd multiples of fsample/2 that depend on offsets, and those offsets may also drift with temperature.
You could insert a low-pass filter with a cut-off frequency far in the ultrasonic region between ADC and DAC to see if that reduces the spurs. You could also touch a pin of a crystal (assuming it's accessible and there are no dangerous voltages on it) to see if that shifts the frequencies.
Thank your for your answering and for you thoughts.
These old but faithful and robust RME Multiface 1 are fitted with AK4528VF codec's, housing both the DA- and AD-Sections within one chip. So there is always one and the same clock signal active for both DA and AD sections at a time.
There are indeed two (always active) oscillators inside the box. One is for the 44.1kHz/88.2kHz, and the other for the 48kHz/96kHz sampling rates, and the one or the other is selected/connected at time to the codec's, depending on the selected sampling rate. So there is no mixing two dystuned oscillators for the proper AD- and the DA-processes.
But both oscillators are active while the multiface is powered. So maybe there is still the theoretical possibility of induced and drifting interferences, which could e.g. spread through the power supply lines? It's rather unbeleivable ... But I certainly will track this issue down by isolating the supplies of the oscillators, if no other, stronger theories emerge from this call for help.
Then again, considering the amount of shift you will notice that this shift spans over a > 3:1 section which by far exceeds a crystal's temp. drift over time. Typically values for this drift would enumerate in ppm's.
Therefore, for the moment, the conceptual search for the real cause might go on ... So I am still ready collecting good ideas before turning on my welding station. But meanwhile I will certainly touch both of the oscillators with my fingers, severel times, as you suggest ... and alternatively tract them with crushed ice also.
Last edited:
The shifts I see are in the range of 1 kHz to 4 kHz. Assuming you have 22.5792 MHz and 24.576 MHz oscillators, their frequencies are 3 * 7^2 * 153600 Hz and 2^5 * 5 * 153600 Hz and their smallest common multiple is 2^5 * 3 * 5 * 7^2 * 153600 Hz = 3.612672 GHz, this being the 160th harmonic of 22.5792 MHz and the 147th of 24.576 MHz.
So when the 160th harmonic of one crystal frequency mixes with the 147th of the other, you only need drifts of the order of 1 kHz/3.612672 GHz ~= 0.2768 ppm to 4 kHz/3.612672 GHz ~= 1.1072 ppm to explain the drift of the product in the audio band. Those harmonic numbers are pretty high, though, but we are talking about quite small interferences.
The other hypothesis was ADC idle tones. That's actually a bit unlikely, as those should mainly show up during silence and you measure with a very large signal. Can you add a small offset somehow to see if that makes the tones move?
So when the 160th harmonic of one crystal frequency mixes with the 147th of the other, you only need drifts of the order of 1 kHz/3.612672 GHz ~= 0.2768 ppm to 4 kHz/3.612672 GHz ~= 1.1072 ppm to explain the drift of the product in the audio band. Those harmonic numbers are pretty high, though, but we are talking about quite small interferences.
The other hypothesis was ADC idle tones. That's actually a bit unlikely, as those should mainly show up during silence and you measure with a very large signal. Can you add a small offset somehow to see if that makes the tones move?
Hi Marcel
Thx! I will track your ideas further down "in situ" tomorrow. You will hear/read from the results.
Regards
Simon
Thx! I will track your ideas further down "in situ" tomorrow. You will hear/read from the results.
Regards
Simon
Cooling / Heating / Touching the oscillators does not make any difference. Cooling the LM2585S step-up flyback voltage converter (f_switch = 100kHz) instead partially reverts the thermal drifting. There is also a LM2595 (f_switch = 150kHz) inside the box. In the data sheet, both of them show a nicely fitting f_switch_drift of 3kHz between 0°C and 50°C. Therefore it seems obvious that the cause of these drifting spikes is to be located there.
It's a pity neither the LM2585S nor the LM2595 can be synchronized to a multiple of the sample rate. Any possibilities to shield or to replace inductors with an open magnetic system with closed ones?
Thank you for the most valuable input!
First measure: Shielding a little bit better the power supply section with a neat piece of iron foil cut out of a sweeties can. The sacrified sweeties by themselves were delicious.
Second measure: Buffering three supply points with polymer capacitors. These points were the Vcc of the oscillators supply (because of small ripple there seen on the oscilloscope), and Vref and Va of the AK4528 codec (by trial and error).
The result is quite pleasant, but only partially: There are no more migrating spikes now. There is one remaining, stable and annoying spike @ 22.5kHz. That is SR/4 at 88.2kHz. Unfortunately, I could not get rid of that one.
First measure: Shielding a little bit better the power supply section with a neat piece of iron foil cut out of a sweeties can. The sacrified sweeties by themselves were delicious.
Second measure: Buffering three supply points with polymer capacitors. These points were the Vcc of the oscillators supply (because of small ripple there seen on the oscilloscope), and Vref and Va of the AK4528 codec (by trial and error).
The result is quite pleasant, but only partially: There are no more migrating spikes now. There is one remaining, stable and annoying spike @ 22.5kHz. That is SR/4 at 88.2kHz. Unfortunately, I could not get rid of that one.
