DSD to PCM Conversion A/B Test Results

[Aries Cerat]

I don’t think there are many recordings that have more than 18 bits of resolution.

Hi

You go the point correct, but..
It is actually worse.
There is no recording ever made with more than 15 bits of resolution, true resolution. The dynamic range of the mic-mic pre chain cannot transcend this barrier. Rest is random noise.(even for a 16bit rec, the 14-15 bit is actual useful data)

The average recordings are actually 14 bits at max.

Our dacs shift the 16 bits to 18 bits (so useful resolution is the native 16 bit ) and on 24bit you use the 18 MSBs.

24bit and 32 bit audio is a good marketing tool, but these is the state of things unfortunately. Actual data on a 24 bit recording is the first 14-15 bit, rest is random noise.

Cheers
Stavros

 

[Argonaut]

The Aries Cerat Kassandra Limited Edition DAC used for your tests has an AD1865 ADC chip. It is an 18-bit part. This means that any and all data sent to the Kassandra Limited Edition DAC is converted to no more than 18-bits at it's analog output. Since this is not an audio standard I am assuming it is actually 16-bits being sent to the AD1865. So, regardless of the data you sent the Aries Cerat Kassandra Limited Edition DAC you only hear 16-bits from it.

It looks like you tested the capability of the Aries Cerat Kassandra Limited Edition DAC to convert files down to 16-bit PCM

Sorry … To which DAC are you referring

 

[Believe High Fidelity]

Here is a visual representation of how the conversion process was done.

 

[Al M.]

Hi

You go the point correct, but..
It is actually worse.
There is no recording ever made with more than 15 bits of resolution, true resolution. The dynamic range of the mic-mic pre chain cannot transcend this barrier. Rest is random noise.(even for a 16bit rec, the 14-15 bit is actual useful data)

The average recordings are actually 14 bits at max.

Our dacs shift the 16 bits to 18 bits (so useful resolution is the native 16 bit ) and on 24bit you use the 18 MSBs.

24bit and 32 bit audio is a good marketing tool, but these is the state of things unfortunately. Actual data on a 24 bit recording is the first 14-15 bit, rest is random noise.

Cheers
Stavros

Interesting, thanks. On the playback side there appear to be similar limitations. While I am not sure if that is correct, I have read that the average stereo system has an effective dynamic range of about 80 dB (less than 14 bit resolution, if one bit is 6 dB dynamic range or 6 dB signal/noise ratio).

 

[Ian B]

I think there are a few issues with the test and discussion as others have noted:

1. It is not possible to record at DSD512. The test track is 8x upsampled DSD64, and then downsampled

2. 32 bit is not supported by most audiophile DACs. 32 bit comes in two versions float point and integer. 32 bit float point is just a container for a 24 bit file, and a way for audio engineers to avoid digital clipping. 32 bit integer is the true 32 bit file, although the output of the ADC chip is the same as 24 bit. The only true difference is in precision of the ADC and DAC filters. A couple professional converters support this format, and it is not common in hifi. Michal Jurewicz of Mytek is a proponent.

3.The equivalency of digital bits and analog signal:

I worked for a few years in a studio that ran Studer A827 tape machines along side of ProTools, and have had my own digital recording rig for about 15 years. 24 bit was clearly more accurate to both the microphone source as well as the tape source than 16 bit. 16 bit typical was a pretty inadequate capture medium, even for tape tracks. The "formula" says that tape should be 13 bits, which is bonkers in the real world and obviously wrong.

We did have some 80s equipment that operated between 12 and 14 bit, and was neither sonically equivalent with 16 bit or 24 bit. 15 bit does sound about the same as 16, bit though (and you can hear on most HDCDS if you don't have an encoder). Likewise most SACD DSD64 performance should be the equivalent of 18 bit, but they sound nothing alike. DSD, and to a lesser extend 24/192 PCM were the lowest rates needed to do justice to 2" 24 track tape, let alone a microphone input.

So generally discussions of "how many bits" constitute music, or are equivalent with analog signal chain are misguided . You can dither a 24 bit recording down to 16 bit or lower (with any type of dither) and hear the quality loss easily enough. Digital and analog signals are not apples to apples.

Likewise, we all know that it is possible to build a pleasing and well regarded 18 or 16 bit DAC that might sound better to a listener than a by-the-book, 24 bit capable DAC playing a 24 bit file. But from the signal perspective of the original source, this is inadequate, and the conversions are not totally transparent.

 

[Aries Cerat]

Interesting, thanks. On the playback side there appear to be similar limitations. While I am not sure if that is correct, I have read that the average stereo system has an effective dynamic range of about 80 dB (less than 14 bit resolution, if one bit is 6 dB dynamic range or 6 dB signal/noise ratio).

This is also correct.There is an exact mathematical analogy between the dynamic range and bit equivalent. If the noise floor of a room is 30db and the max undistorted peak output is 110db then yes it is 80db or equivalent of about 14 bit.

You cannot record in 24 bit and expect that after the 14-15th bit to contain actual information and not random (or even deterministic noise).If you convert a file from 24bit to 16 (not with just by trunking conversion )and hear differences then it is a product of the conversion and not actual difference due to more usefull info/data after the 15th bit.

 

[microstrip]

Hi

You go the point correct, but..
It is actually worse.
There is no recording ever made with more than 15 bits of resolution, true resolution. The dynamic range of the mic-mic pre chain cannot transcend this barrier. Rest is random noise.(even for a 16bit rec, the 14-15 bit is actual useful data) (...)

Do you have any reference that supports this point of view? For example HDCD recordings encoded the lower bits 17-20 in bit 16 and sounded different from common 16 bit recordings. It is know that we can listen to signals buried in noise.

Should we assume that when we praise our DXD recordings we are being fooled by bias expectation?

 

[Aries Cerat]

Do you have any reference that supports this point of view? For example HDCD recordings encoded the lower bits 17-20 in bit 16 and sounded different from common 16 bit recordings. It is know that we can listen to signals buried in noise.

Should we assume that when we praise our DXD recordings we are being fooled by bias expectation?

Do not confuse higher data rate to higher bit word. Higher data rate does provide more data/info. Higher bit word ,after one point,does not.DXD sounds better due to vast superior info due to higher data rate.(sampling rate)

About what you asked.you can replicate. Since you want references you can get a first hand data if you care to make the experiment.

You have two recordings,one 16bit and one 24 bit both containing true white noise(not computer false-random generated) but actual white noise.
Inject a signal at -110db to the 24 db recording.

After that try to isolate back the signal of -110db or ,try to differentiate the two signals based on the injected signal.
Guess what the result is.

Not sure how you ended with your end statement.

 

[microstrip]

Do not confuse higher data rate to higher bit word. Higher data rate does provide more data/info. Higher bit word ,after one point,does not.DXD sounds better due to vast superior info due to higher data rate.(sampling rate)

Ok, nice to know you consider that higher bit rates sound better.

About what you asked.you can replicate. Since you want references you can get a first hand data if you care to make the experiment.

You have two recordings,one 16bit and one 24 bit both containing true white noise(not computer false-random generated) but actual white noise.
Inject a signal at -110db to the 24 db recording.

After that try to isolate back the signal of -110db or ,try to differentiate the two signals based on the injected signal.
Guess what the result is.

Not sure how you ended with your end statement.

My apologies, I am not interested in such experiences that I will not carry. Long ago I found that listening to a 44.1 kHz 24 bit recording sounded different - in my opinion better - than the same 16 bit CD. Since you have no references to substantiate your claim I will stay with AES recommendations - "High-quality Analog to Digital conversion should be used to convert Analog Masters to Digital Backups/Safeties in either PCM or DSD form using sample rates of at least 88kHz and 24bits or greater depth. "

 

[Aries Cerat]

We agree to disagree.
Btw the recommendation does not equal any other than a recommendation.I would like to accommodate you in a test that i described above.you see we do not follow vague recommendations but sometimes walk the talk to figure things on ourselves.
I would be happy to invite you to a test where you would asked to differentiate a -110db signal over -80db noise floor by ears or any electrical instrument.

Anyway maybe too heavy discussion for the forum,but happy to discuss even if we disagree 180deg

 

[Ian B]

Not sure if he'd want to get involved, but you could always ask Bruce. His work literally involved authoring 16 and 24 bit files at various sample rates from DSD masters.

 

[Bruce B]

Guess they were only interested in making a SACD at the time. What has changed since you did your DSD Battle thread, what is now the best A to D?

Still say the Merging Horus is tops!

 

[Bruce B]

Guess they were only interested in making a SACD at the time. What has changed since you did your DSD Battle thread, what is now the best A to D?

IMHO.... an SACD sounds better recorded at 2xDSD.... even after the conversion!

 

[microstrip]

We agree to disagree.
Btw the recommendation does not equal any other than a recommendation.I would like to accommodate you in a test that i described above.you see we do not follow vague recommendations but sometimes walk the talk to figure things on ourselves.
I would be happy to invite you to a test where you would asked to differentiate a -110db signal over -80db noise floor by ears or any electrical instrument.

Anyway maybe too heavy discussion for the forum,but happy to discuss even if we disagree 180deg

Thanks. If we play a 44.1 kHz / 24 bit file on your DAC you discard the 8 less significative bits?

 

[Aries Cerat]

Thanks. If we play a 44.1 kHz / 24 bit file on your DAC you discard the 8 less significative bits?

The six lsb

 

[Lee]

I can add an observation from the recording side of things...

Nick Arroyo and I do live to two track recordings of the Peachtree String Quartet and other Atlanta Symphony musicians. We have been doing this since the late 90s and use two AKC 414-BULS mics in ORTF configuration. We split the mic feed into both a SoundDevices PCM box and a Korg DSD box. We run everything off batteries as we discovered that the local churches we record in have very noisy power.

Both of us and several of our musician friends hear DSD capture the live event better.

We have replicated this results with several ADCs and several recording boxes. The DSD playback is simply more natural.

Also, we have done many live to two track violin and cello solo recordings. Only 24/88 or higher captures the string timbre properly when using PCM. Hirez digital really does make a difference.

 

[microstrip]

I can add an observation from the recording side of things...

Nick Arroyo and I do live to two track recordings of the Peachtree String Quartet and other Atlanta Symphony musicians. We have been doing this since the late 90s and use two AKC 414-BULS mics in ORTF configuration. We split the mic feed into both a SoundDevices PCM box and a Korg DSD box. We run everything off batteries as we discovered that the local churches we record in have very noisy power.

Both of us and several of our musician friends hear DSD capture the live event better.

We have replicated this results with several ADCs and several recording boxes. The DSD playback is simply more natural.

Also, we have done many live to two track violin and cello solo recordings. Only 24/88 or higher captures the string timbre properly when using PCM. Hirez digital really does make a difference.

When looking at such subjects we must separate the recording format from the distribution format. It is known that when recording using an high rate the anti aliasing filters can have lower slope and the system is less susceptible to noise ultrasonic or transients. A 44.1/16 consumer file does not imply that the recording was carried at this bit rate.

In fact, current modern recording systems are much more complex than people imagine - a DXD recoding does not mean that the recording was carried in pure PCM 382/24 - such thing does not exist!

 

[Hear Here]

Can I ask why you should want to convert DSD files to PCM.

I have never used DSD as Qobuz offers only PCM and my ripped CD Library is all PCM, but DSD advocates clain it is superior to PCM. So, if that's the case (it msy not be), why convert to an inferior spec, rather than use a DAC that can accpt DSD?

 

[Lee]

Can I ask why you should want to convert DSD files to PCM.

I have never used DSD as Qobuz offers only PCM and my ripped CD Library is all PCM, but DSD advocates clain it is superior to PCM. So, if that's the case (it msy not be), why convert to an inferior spec, rather than use a DAC that can accpt DSD?

Many DACs accept DSD and I tend to prefer native DSD where available. That said, with a good quality DSD to high rate PCM conversion, the sound will still be excellent.

 

[Rexp]

I can add an observation from the recording side of things...

Nick Arroyo and I do live to two track recordings of the Peachtree String Quartet and other Atlanta Symphony musicians. We have been doing this since the late 90s and use two AKC 414-BULS mics in ORTF configuration. We split the mic feed into both a SoundDevices PCM box and a Korg DSD box. We run everything off batteries as we discovered that the local churches we record in have very noisy power.

Both of us and several of our musician friends hear DSD capture the live event better.

We have replicated this results with several ADCs and several recording boxes. The DSD playback is simply more natural.

Also, we have done many live to two track violin and cello solo recordings. Only 24/88 or higher captures the string timbre properly when using PCM. Hirez digital really does make a difference.

If DSD is better, why use PCM at all?