Why doesn't DSD sound as "analog" as it used to?

This is me theorizing, but also curious if anyone with more technical expertise can provide some insight.

I first got into DSD in the early 2000s with a Pioneer DV-45a disc player and a small collection of SACDs. I had just started interning at analog capable recording studios and was really impressed that my budget system sounded so much like what I was hearing in the studio. Eventually one studio I was working at as an assistant purchased a Tascam DV-RA1000 DSD recorder and that cheap box at 2.8mhz, put our 1/4" stereo machine and digital stereo recorders out to pasture. Years later, I bought my own Korg MR-1000 and Tascam DA-3000, double rate capable DSD recorders, and despite their cheap build, tape and vinyl transfers sounded extremely authentic.

In those days the main gripe I had with DSD of all sorts was that the dynamic range was a bit compressed, and the high frequencies always seemed a bit dark and smeared off. But everything else was great, and the tradeoff was not so bad at DSD128.

However, with modern DSD capable DACs, and ADCs, the sound I hear reminds me more of PCM in some key ways. It started in an obvious way with the Merging Horus professional system, and then consumer ESS chip DACs that convert 1 bit DSD to 6 bit. But now most of the FPGA single bit DACs I’ve heard do it as well.

Of DSD capable DACs I’ve owned recently, Mytek Manhattan II (ESS9038), PS Audio DSS, and EMM Labs DA2, each sounded a bit less direct and three dimensional than the cheap old Burr-Brown chip devices I was used to, but perform better. This surprised me most with the EMM Labs because my most coveted SACDs were all mastered with their ADCs. Bucking the trend, some multibit DSD schemes, like DCS’ 5 bit version actually sound excellent, although I’m not a fan of their PCM or upsampling.

The old tradeoff has balanced though, DSD and DSD upsampling no longer darkens or smears the top end, and the dynamic range has improved. More hifi, high performance attributes, but less “straight from tape” character. Modern DSD DACs are ultra smooth, low fatigue, detail machines.

My theory is in the hifi market, that loss of high frequency is unacceptable, so newer DSD designs use a lot more noise shaping (negative feedback to increase audio band dynamic range). This seems especially important for PCM-DSD upsampling, where the customer would be disappointed if it seemed like their old collection actually lost something in the DSD conversion. But like any negative feedback, some of the raw and 3D character is lost.

It seems a bit analogous to NOS DACs in the PCM world: they perform objectively worse, but losing the digital filter sounds subjectively more “analog” and raw.


Have you gone back to listen to a DV-RA1000 (for example) to see if it still sounds a bit more direct and three dimensional as you recall? My audio memory regularly plays tricks on me over time.

MY audio failings aside, my guess is what you are hearing is a difference in filters.

Your comments also remind me of those who absolutely prefer PCM, finding it to have bite and energy they enjoy.

1 Like

I haven’t listened to the DV-RA1000 for many years, but the DA-3000 still sounds good in this regard, if a little bit grungy compared to higher end DACs. I actually put it next to a Mytek Brooklyn ADC (AKM-based) recording an analog mix at DSD128, and the Tascam sounded more like the analog while the Mytek was more detailed and had more accurate highs, but PCM-like. The same thing happened when we sent mixes recorded on the DA-3000 to a mastering engineer who used the Merging Horus, more detail, less depth, more PCM-like.

Filters can definitely play a role. I think especially so in PCM-DSD upsampling. I used to use Korg Audiogate software to convert PCM to DSD in conjunction with the Korg or Tascam DSD recorders, and I always found that the result had a very analog output, but was sort of “DSD-colored” like it changed the sound in a more euphonic way. Audirvana’s SOX upsampler did that to a lesser degree when using the 5th order noise shaper, but HQPlayer doesn’t sound like that at all with any filters (always a bit PCM-like, with all but the most relaxed minimum phase filters), and the upsampling DACs fall more in with the latter two being very clean and flat in comparison.

I reasoned with some of the upsamplers that a lot of them operate at DXD (352/384) and don’t use the full DSD bandwidth, but some of it is still mysterious to me. Some of the old upsamplers might be using minimum phase filters too.

1 Like

I use a DA-3000 quite a bit, but relatively little for DSD as few people want a recording in DSD. But the DSD side is appealing. I understand what you like about it.

I find the unit remarkably faithful recording whatever is fed to it, recording PCM or DSD.

1 Like

I love it. It is fantastic at DS128, but surprisingly good in PCM as well, esp 24/192.

1 Like

One thing I am remembering too about the ancient SACD players: The original spec for SACD (Scarlett Book) was 5th order noise shaping, and sometimes even 3rd order. I’m sure that newer DSD DACs are using 8th or higher, and the most popular modulators on HQPlayer are 7th order, I think. This is a big increase in noise shaping.

Also, the earliest players would allow bandwidth all the way up to 100khz as part of the Scarlett Book spec, but later added extra 50khz filters to prevent the SACD ultrasonic noise from damaging super tweeters, etc. So more analog filtering too. I think the DSD recorders mentioned above did maintain the higher bandwidth however.

Quoted from Roon:
The 24kHz filter results in frequency response characteristics that are nearly guaranteed not to offend your PCM gear, but this involves placing a filter fairly near to the limits of human frequency perception. It’s a conservative/safe setting.

The 30kHz filter is the best compromise, it’s the default in Roon, and what we recommend for most users. This setting nicely removes nearly all of the noise spectrum in DSD, while leaving some space between stuff we can hear and the filter itself.

The 50kHz filter preserves significantly more noise (especially during DSD64 playback), but is even less likely than the 30kHz filter to muck with frequency response in an audible way. This doesn’t mean it’s “better”. It’s possible that with some gear, this filter will sound worse than 30k depending on how the gear reacts to the HF noise.

This adds another variable as Roon is adding a digital filter to a DSD bitstream. The only way they can do this is with a multibit DSD DSP like ESS uses in their DAC chips, or Ted uses in his volume control, or by converting DSD to PCM. Any of these filters have the possibility of degrading the SQ, and it’s probably most benign to do this in the analog realm.

The Roon filters mentioned above are only used when converting DSD files to PCM for playback through systems that don’t have DSD support. Digital filters are literally the only way this can be done – to do it in analog you’d have to have a DSD-capable DAC… in which case you wouldn’t be converting to PCM.

Coincidentally, the basic concept of the Ted’s DAC is that it converts DSD bits directly into +/- voltage and then runs that through an analog low-pass filter. So yeah, an analog LPF for DSD is a good idea in the right place.

1 Like

Does Roon not use those the other way around when converting from PCM to DSD to control noise, or that the role of the 5th and 7th order filters?

Any PCM to DSD upsampler will use filters, but these aren’t to reduce DSD noise, which is actually impossible. They exist to prevent PCM aliasing distortion. Anything that is played back from a DAC as DSD will produce noise at the output that can’t be removed with a digital filter. There are only a few ways to deal with it: Analog filters, noise shaping, conversion to multibit DSD, and upsampling.

The 5th or 7th order noise shapers aren’t proper filters, but are basically a feedback loop that moves the noise that DSD produces out of the audio range. The more aggressive the feedback, the less noise in the audio band (greater dynamic range), and the noise moves farther above it.

Upsampling DSD to a higher frequency does a similar thing: by doubling the DSD frequency you can also double the frequency where the noise starts to build up, pushing it farther away.

And the farther the noise is from the audio band, the mellower and better sounding your analog output filter can be. Using really steep analog filters causes phase distortion and some other nasties like in the earliest CD players.

Some DACs also will take the single bit of DSD and “remodulate” it to 5 or 6 bit without converting it to PCM. Having more bits allows less noise and better performance, but you lose any advantage of 1 bit conversion, and there will be at least a little quality lost in the remodulation. That’s what ESS and DCS do and a few others.

Any of these things in excess can produce a more “processed” sound.

1 Like

Thanks Ian. This is good to know.

Oversampling & PCM to DSD never worked for me. LPs still do a good job here.

Not a DS DAC owner, S?

Hi @badbeef
I am aware that my DS DAC does that conversion, but not like using a PC or Nucleus. Theo does it differently.

Ah - I see. Thought you were saying you were analog-only. Yeah - hardware tends to do it better than software. Pretty computationally intensive.

Yes, I was hoping to get closer to Analogue, but I realized it was an illusion :woozy_face: So for analogue, I have to play analogue.

1 Like