A concern is often expressed about the effect of Direct Stream Digital (DSD) high frequency (HF) quantization noise on downstream electronics, especially power amplifiers, and the tweeters of loudspeakers.
For a quarter of century since the inception of SA-CD, the first carrier of DSD, I have never seen or heard about a single properly documented case of a piece of electronic or speaker unit disturbed, let alone damaged, by the HF shaped quantization noise of DSD signals. Except one possible case which was publicly acknowledged by Mr Andrew Demery, then in charge at Philips for assisting the deployment of DSD and SA-CD in the pro market.
Despite the widespread concerns expressed by some people about DSD HF quantization noise in the Hifi press and on the Web, I have never stumbled across a properly conducted experience designed to assess possible effects of said noise on the behaviour of downstream electronics or speakers, except one published in the British magazine Hi-Fi News and Record Review (more of that latter).
Therefore, I propose to tackle this issue.
I have no access to DSD test signals of any kind save the very limited Denon Audio Check SA-CD (reference COGQ-28), which is intended to help tweaking an actual stereo or multichannel loudspeaker installation, not to test electronics or loudspeakers. I have no high performance test equipment either.
Nevertheless, I propose to compare DSD HF noise level to the HF switching noise level of amplifiers using the now popular class D technique.
To do this, I will start to point out to a relevant post by our kind host Amirm, who made comments about the noise-shaped quantization noise level of a pro-audio interface, the Focusrite Scarlett 2i2 Gen3. Amirm not only provided a relatively wide-band FFT analysis of the frequency spectrum of this interface when converting PCM signal in order to visualize the shape of the high-frequency noise spectrum, but he also measured the actual RMS voltage level of this noise inside a 90 kHz bandwidth.
Having no Audio Precision, I made some measurements of the same kind as Amirm with more modest analogue test equipment. I used a laboratory-grade active high pass filter with a sufficiently high frequency bandwidth (10 MHz @ -3 dB) to remove the low frequency content at the output of SA-CD players in order to only keep the part of the spectrum above a set corner frequency. Then, I measured the RMS noise level at the output of the active filter with a psophometer (a kind of voltmeter especially designed to measure noise).
As I own a player (Marantz DV-12S2) that uses the same DAC chip or at least of the same family as those used in the above mentioned Focuriste interface, I have measured first the shaped quantization noise level at the output of this player when it plays a signal recorded in PCM on a test disc. Here are the results:
Overall, the result seems to me consistent with those published by Amirm. Obviously, most of the noise is above 50 kHz, as shown in Armim's FFT analysis of the Focusrite.
As I consider that this first results validate the methodology, I have measured the same way the HF noise level of all the operational SA-CD players I have at hand:
The slightly higher noise level in the 20 to 300 kHz bandwidth measured with some players is most probably due to leakage in the pass band of signal content in the stop band, because the active filter I used has a relatively low rate of attenuation.
I have checked the noise levels with a Leader LMV-181A AC voltmeter that has a bandwidth of 1 MHz @ -3 dB. Despite the wider bandwidth of this voltmeter, the noise level readings are actually slightly lower than with the psophometer. I think that is due to the better detector circuit of the psophometer, which is capable of converting noise-like signal having high crest-factor (peak to average level) in more accurate RMS values. The point is that we can deduce from the measurements made with the Leader voltmeter that the quantization noise levels at the output of each SA-CD player have a significant frequency content mostly above 50 kHz and well under 300 kHz or so. This is consistent with the wide-band frequency response plot of typical SA-CD players most of us must be familiar with.
The Marantz player obviously has the widest bandwidth and lets much more quantization noise leaks at its output. Consequently, I decided to pursue the experience with this particular player considering it as a worst-case.
I have measured the noise level at the output of my preamp with its volume control set at the usual position to listen to well recorded classical music discs. Here are the results, in more bandwidths than before in order to show the approximate frequency above which the player-preamp combination begin to roll off the noise:
There is no surprise: the measured levels are about 24 dB below the volume control position that corresponds to unity gain.
Now, I have not yet a power load at my disposal to measure directly at the output of my power amplifier. But it is possible to calculate the worst-case noise level after amplification by multiplying the preamp output noise voltage by the voltage gain of the power amplifier.
I have set the input level of my amplifier (Cabasse Polaris AM1000) such as an input RMS voltage equivalent to 0 dBFS or 0 dB SA-CD at the output of the player (that is, with the overall gain of the preamp set at unity) does produce an output voltage of about 29 V RMS (210 W/4 ohms). That corresponds to a voltage gain of 15 with unbalanced signal. That way, the amplifiers still have about 4.65 dB of peak voltage level headroom above 0 dB SA-CD or 0 dBFS for dealing with over-modulated DSD signals or inter-sample PCM overs.
The worst case noise level at the output of the power amplifier at usual listening level should thus be 15x6.7 mV equal to 100.5 mV.
I think I can confidently speak about worst case noise level because I will make two further observations.
Firstly, it is well known that a DSD signal is a constant power signal. The more power in the signal pass band, the less power in the noise pass band. I have also measured RMS noise levels at the preamplifier output when playing the musical tracks on the Denon SA-CD test disc: the HF noise levels did varied between at least 4.8 mV and at most 6.5 mV in a 50 to 300 kHz bandwidth depending on the specific track. The pink noise track did produce an HF noise level of 5.2 mV in the same bandwidth. From this results, we can say that the HF noise level with a -16 dB SA-CD sinus signal is representative of the worst case noise level that a musical recording will produce at the output.
Secondly, I have simulated the frequency response of the input stage of the power amplifier with the Caneda software. This stage limits the frequency response of the overall power amplifier (not taking into account an input RFI filter whose action is visible from about 8 to 10 MHz). This stage brings a further attenuation of -1.6 dB@50 kHz, -2.7 dB@70 kHz, -3.9 dB@90kHz and -5.5 dB@ 100 kHz. I should point out that the rate of attenuation of the power amplifier input stage frequency response above 100 kHz is much less than the rate of change of the simulated frequency response of the Marantz DV-12S2 output analogue low pass filter. This fact alone made me confident that the power amplifier will cope with the HF noise level produced by said Marantz.
At this point of the experience, I cannot say that the measured HF noise levels have no effects whatsoever on the electronics (pre or power amp) or the loudspeakers. But I find useful to compare the computed HF noise levels at power amplifier output with those measured at the output of some class D power amplifiers. I am perfectly aware that is not comparing like with like, because the low pass filtered DSD HF quantization noise is a broadband noise signal, whereas the low pass filtered switching noise of class D amplifier resembles more a narrow band signal or even single frequency signals. Nevertheless, the comparison could be enlightening.
For comparison, I rely on the data published by Stereophile. There are all public data found on the website of the magazine, so feel free to check them. Bear in mind they are measured with no input signal. To the best of my knowledge, when actually reproducing signals, the switching noise of some amplifiers can be higher and/or occupy a larger frequency span depending on their specific topologies.
At this point, I can only add a summary of an experiment realized by Mr Keith Howard and published in the May 2002 issue of Hi-Fi News (Reference 1).
Mr Howard has recorded the DSD quantization noise produced by a Philips SACD-1000 SA-CD player playing the 1kHz@-160 dB SA-CD signal of the Philips Super Audio CD DAC Test Disc (ref. 3122-783-00632) with its analogue output filter set to the higher cutoff frequency. This noise was digitally analyzed to filter a white noise produced by the Cool Edit Pro software to obtain a noise spectrum at 192 kHz sample rate resembling DSD HF noise. Then, the resulting noise was mixed with some 24 bits/96 kHz musical excerpts up-sampled to 192 kHz. M. Howard then used a replay software to blind test the WAV files containing the altered version of each music tracks against the original (also previously up-sampled at 192 kHz). The replay chain consisted of a LynxTWO sound card, a potentiometer-based passive preamp and a pair of Exposure XVIII mono power amplifiers driving B&W CDM1NTs supplemented by prototype ribbon super-tweeters.
The conclusion of the blind test that Mr Howard has made on himself are : "In fact I pretty quickly abandoned attempts to achieve a significant score on ABX blind testing as a hopeless cause. Whatever the differences were, they were too subtle to secure a statistically significant result in just eight trials - and how anybody could remain sane for more than eight trials I can't imagine." Nevertheless, M. Howard claims he was able to identify a subtle difference in sighted listening, but that was to be expected in an article published in a Hi-fi magazine.
That's about as much experimentation I am aware of concerning the subject of this topic.
Reference :
1. Keith Howard, Noises off, Hi-Fi News and Record Review, Vol 47 No 5, May 2002, page 78.
For a quarter of century since the inception of SA-CD, the first carrier of DSD, I have never seen or heard about a single properly documented case of a piece of electronic or speaker unit disturbed, let alone damaged, by the HF shaped quantization noise of DSD signals. Except one possible case which was publicly acknowledged by Mr Andrew Demery, then in charge at Philips for assisting the deployment of DSD and SA-CD in the pro market.
Despite the widespread concerns expressed by some people about DSD HF quantization noise in the Hifi press and on the Web, I have never stumbled across a properly conducted experience designed to assess possible effects of said noise on the behaviour of downstream electronics or speakers, except one published in the British magazine Hi-Fi News and Record Review (more of that latter).
Therefore, I propose to tackle this issue.
I have no access to DSD test signals of any kind save the very limited Denon Audio Check SA-CD (reference COGQ-28), which is intended to help tweaking an actual stereo or multichannel loudspeaker installation, not to test electronics or loudspeakers. I have no high performance test equipment either.
Nevertheless, I propose to compare DSD HF noise level to the HF switching noise level of amplifiers using the now popular class D technique.
To do this, I will start to point out to a relevant post by our kind host Amirm, who made comments about the noise-shaped quantization noise level of a pro-audio interface, the Focusrite Scarlett 2i2 Gen3. Amirm not only provided a relatively wide-band FFT analysis of the frequency spectrum of this interface when converting PCM signal in order to visualize the shape of the high-frequency noise spectrum, but he also measured the actual RMS voltage level of this noise inside a 90 kHz bandwidth.
Having no Audio Precision, I made some measurements of the same kind as Amirm with more modest analogue test equipment. I used a laboratory-grade active high pass filter with a sufficiently high frequency bandwidth (10 MHz @ -3 dB) to remove the low frequency content at the output of SA-CD players in order to only keep the part of the spectrum above a set corner frequency. Then, I measured the RMS noise level at the output of the active filter with a psophometer (a kind of voltmeter especially designed to measure noise).
As I own a player (Marantz DV-12S2) that uses the same DAC chip or at least of the same family as those used in the above mentioned Focuriste interface, I have measured first the shaped quantization noise level at the output of this player when it plays a signal recorded in PCM on a test disc. Here are the results:
Overall, the result seems to me consistent with those published by Amirm. Obviously, most of the noise is above 50 kHz, as shown in Armim's FFT analysis of the Focusrite.
As I consider that this first results validate the methodology, I have measured the same way the HF noise level of all the operational SA-CD players I have at hand:
The slightly higher noise level in the 20 to 300 kHz bandwidth measured with some players is most probably due to leakage in the pass band of signal content in the stop band, because the active filter I used has a relatively low rate of attenuation.
I have checked the noise levels with a Leader LMV-181A AC voltmeter that has a bandwidth of 1 MHz @ -3 dB. Despite the wider bandwidth of this voltmeter, the noise level readings are actually slightly lower than with the psophometer. I think that is due to the better detector circuit of the psophometer, which is capable of converting noise-like signal having high crest-factor (peak to average level) in more accurate RMS values. The point is that we can deduce from the measurements made with the Leader voltmeter that the quantization noise levels at the output of each SA-CD player have a significant frequency content mostly above 50 kHz and well under 300 kHz or so. This is consistent with the wide-band frequency response plot of typical SA-CD players most of us must be familiar with.
The Marantz player obviously has the widest bandwidth and lets much more quantization noise leaks at its output. Consequently, I decided to pursue the experience with this particular player considering it as a worst-case.
I have measured the noise level at the output of my preamp with its volume control set at the usual position to listen to well recorded classical music discs. Here are the results, in more bandwidths than before in order to show the approximate frequency above which the player-preamp combination begin to roll off the noise:
There is no surprise: the measured levels are about 24 dB below the volume control position that corresponds to unity gain.
Now, I have not yet a power load at my disposal to measure directly at the output of my power amplifier. But it is possible to calculate the worst-case noise level after amplification by multiplying the preamp output noise voltage by the voltage gain of the power amplifier.
I have set the input level of my amplifier (Cabasse Polaris AM1000) such as an input RMS voltage equivalent to 0 dBFS or 0 dB SA-CD at the output of the player (that is, with the overall gain of the preamp set at unity) does produce an output voltage of about 29 V RMS (210 W/4 ohms). That corresponds to a voltage gain of 15 with unbalanced signal. That way, the amplifiers still have about 4.65 dB of peak voltage level headroom above 0 dB SA-CD or 0 dBFS for dealing with over-modulated DSD signals or inter-sample PCM overs.
The worst case noise level at the output of the power amplifier at usual listening level should thus be 15x6.7 mV equal to 100.5 mV.
I think I can confidently speak about worst case noise level because I will make two further observations.
Firstly, it is well known that a DSD signal is a constant power signal. The more power in the signal pass band, the less power in the noise pass band. I have also measured RMS noise levels at the preamplifier output when playing the musical tracks on the Denon SA-CD test disc: the HF noise levels did varied between at least 4.8 mV and at most 6.5 mV in a 50 to 300 kHz bandwidth depending on the specific track. The pink noise track did produce an HF noise level of 5.2 mV in the same bandwidth. From this results, we can say that the HF noise level with a -16 dB SA-CD sinus signal is representative of the worst case noise level that a musical recording will produce at the output.
Secondly, I have simulated the frequency response of the input stage of the power amplifier with the Caneda software. This stage limits the frequency response of the overall power amplifier (not taking into account an input RFI filter whose action is visible from about 8 to 10 MHz). This stage brings a further attenuation of -1.6 dB@50 kHz, -2.7 dB@70 kHz, -3.9 dB@90kHz and -5.5 dB@ 100 kHz. I should point out that the rate of attenuation of the power amplifier input stage frequency response above 100 kHz is much less than the rate of change of the simulated frequency response of the Marantz DV-12S2 output analogue low pass filter. This fact alone made me confident that the power amplifier will cope with the HF noise level produced by said Marantz.
At this point of the experience, I cannot say that the measured HF noise levels have no effects whatsoever on the electronics (pre or power amp) or the loudspeakers. But I find useful to compare the computed HF noise levels at power amplifier output with those measured at the output of some class D power amplifiers. I am perfectly aware that is not comparing like with like, because the low pass filtered DSD HF quantization noise is a broadband noise signal, whereas the low pass filtered switching noise of class D amplifier resembles more a narrow band signal or even single frequency signals. Nevertheless, the comparison could be enlightening.
For comparison, I rely on the data published by Stereophile. There are all public data found on the website of the magazine, so feel free to check them. Bear in mind they are measured with no input signal. To the best of my knowledge, when actually reproducing signals, the switching noise of some amplifiers can be higher and/or occupy a larger frequency span depending on their specific topologies.
At this point, I can only add a summary of an experiment realized by Mr Keith Howard and published in the May 2002 issue of Hi-Fi News (Reference 1).
Mr Howard has recorded the DSD quantization noise produced by a Philips SACD-1000 SA-CD player playing the 1kHz@-160 dB SA-CD signal of the Philips Super Audio CD DAC Test Disc (ref. 3122-783-00632) with its analogue output filter set to the higher cutoff frequency. This noise was digitally analyzed to filter a white noise produced by the Cool Edit Pro software to obtain a noise spectrum at 192 kHz sample rate resembling DSD HF noise. Then, the resulting noise was mixed with some 24 bits/96 kHz musical excerpts up-sampled to 192 kHz. M. Howard then used a replay software to blind test the WAV files containing the altered version of each music tracks against the original (also previously up-sampled at 192 kHz). The replay chain consisted of a LynxTWO sound card, a potentiometer-based passive preamp and a pair of Exposure XVIII mono power amplifiers driving B&W CDM1NTs supplemented by prototype ribbon super-tweeters.
The conclusion of the blind test that Mr Howard has made on himself are : "In fact I pretty quickly abandoned attempts to achieve a significant score on ABX blind testing as a hopeless cause. Whatever the differences were, they were too subtle to secure a statistically significant result in just eight trials - and how anybody could remain sane for more than eight trials I can't imagine." Nevertheless, M. Howard claims he was able to identify a subtle difference in sighted listening, but that was to be expected in an article published in a Hi-fi magazine.
That's about as much experimentation I am aware of concerning the subject of this topic.
Reference :
1. Keith Howard, Noises off, Hi-Fi News and Record Review, Vol 47 No 5, May 2002, page 78.
Last edited:
