High Resolution Audio: Why do we need DSD?

High Resolution Audio - DSD

High resolution digital audio is superb. High sample rates and 24 bit recording is the ultimate conventional way to capture audio, but DSD is said to be better. But is it? Here's our take on it

By far the most accepted and - it has to be said - perfectly functional way to digitise audio is using Pulse Code Modulation (PCM). This is, essentially, drawing sound by numbers. Technically the process is called sampling, and the higher the sample rate, the higher the frequencies that you can reproduce.

There is absolutely nothing wrong with conventional sampling: it does a very good job. If you need higher frequencies then just up the sample rate. If you need more potential dynamic range (the distance between the loud and soft sounds) use 24 bits instead of 16.

With a high sample rate and 24 bits, there is very little lacking from a digital audio recording.

And yet, there are still concerns, especially with very quiet sounds.

Loud sounds are relatively easy. Up around 16 or even 24 bits, there are plenty of "levels" to accurately describe the sounds. But what happens when you have, say, a loud orchestral "crash" followed by a plaintive flute solo? The flute might be extremely quiet compared to the full orchestra. Which means it might only be using a small number of the available bits. This effectively reduces the resolution of its recording. Turn the volume up and you'll hear a sound that's simply not as good as a louder one.

How much of a problem is this? Actually it's not too serious, mainly because if you've just experienced a very loud sound, a much quieter one will be less noticeable, and so, therefore, will be any problems with its quality. But what about when a recording starts very softly? It doesn’t matter that there are louder sounds later in the piece because you will have turned up your volume and will sometimes be able to hear the relatively poor quality caused by encoding a quiet sound with not enough resolution (i.e. not enough bits).

But it's because of this - and also an issue around the quality and repeatability of analogue components in digital to analogue converters - that there are some nagging doubts about PCM recording. It's not that it's not good enough for almost everything, but that there is, perhaps, a better way.

Direct Stream Digital

And that way is, arguably, Direct Stream Digital (DSD).

I don't want to go into too much detail here - some of this is hard stuff to understand if you're not into maths and physics, but the basics are easy enough to understand. Kind of.

With DSD, there's a hugely higher sample rate but only one bit of resolution. While this sounds like a recipe for extreme distortion, actually the opposite is the case. The way this works is rather clever.

When an incoming waveform is sampled in DSD, the result can either be a one or a zero - either full on or full off. Now here's the thing: A high amplitude will result in more ons than offs. A small amplitude will give more offs.

Because of the very high sample rate, it's possible to step back and take the average of all the ons and offs at any point (or, we should say, short time period) and arrive at an almost exact equivalent value for the original waveform at that time. It's this "averaging" that gives DSD a character which is arguably closer to the original analogue signal. It gets rid of "quantization" errors, and, importantly, a quiet signal is reproduced with the same accuracy as a loud one.

Disadvantages? Practically none, as long as you have equipment that can play back DSD. But perhaps the biggest issue is that conventional digital processing doesn't work at all with the format. Virtually all so called DSP (Digital Signal Processing) works on PCM audio. There is no easy way to convert these DSP techniques for DSD. So that means mixing, eq, dynamic range compression and the entire gamut of audio processing methods simply won't work with DSD.

Some DSD enthusiasts claim that the intrinsic difficulty in processing and editing DSD is a guarantee that if they're buying a recording then it is necessarily a direct representation of what actually wen't into the microphone in the first place. Except that it's quite possible that the mastering engineers might have converted the DSD into PCM for processing, and then converted it back, which would, somewhat, defeat the object.

That's OK if all you want to do it listen to it, but it means that it's not going to be an option for anything other than direct recording. A conventional mutitrack studio couldn't function with DSD.

But as a delivery format, it's exceptionally good. Can you hear the difference? It might not be obvious compared to high sample rate PCM, but I suspect that in some cases, there would be a clear difference. And even if it's not immediately apparent, it's good to know that there probably isn't a better way to record and reproduce audio.