For about forty years – since the 1980s – we have been digitizing music. (We are not counting the run-up to the digital age; in fact, there were a lot of trials in this area) Philips and Sony were the pioneers in this field. It is therefore not surprising that they – together – introduced the CD and the spdif (Sony Philips Digital Interface) standard into the market.
Everything digital
When the CD stormed the market in the early 1980s as a replacement for the LP, this went hand in hand with a standard for digital audio. The Redbook standard – part of the Rainbow Book series that defines other standards – defines the requirements for digital audio on a CD. This is crucial if you want to introduce a successful new medium. Without a clear standard you are nowhere.
The CD turned out to be the beginning of thorough digitization. Meanwhile, streaming audio has conquered the hi-fi world and between the CD – which is still popular, just like vinyl for that matter – and streaming audio there have been a few less successful media. Think of DCC and Minidisc. There were even initiatives to distribute music on SD cards. That never got off the ground. As did DRM protected downloads. (Thank goodness!).
Measuring digital audio
Let’s take a look at what we at Alpha Audio look at when we measure a digital device.
For measurements, we primarily use the Prism dScope III. The reason is that it works with fixed scripts, which makes measuring easy and ensures that measurements within Alpha Audio are always comparable. After all, we always use the same settings. The Prism can in fact measure everything that is relevant:
- Linearity – input versus output
- Noise floor – how low is the noise?
- Frequency response – how neutral is the device?
- Phase response – how phase neutral is device / time accurate
- Crosstalk – separation of left and right
- SINAD (Signal Noise and Distortion) – signal noise and distortion in one number
- THD+N – total harmonic distortion with noise
- Imaging (filters) – a visualization of the internal, digital filters
- Dynamic range – maximum output versus noise floor
Above you can see a series of measurements of the Sonnet Pasithea. (And for illustration, a comparison of noise / interference with the MSB Premier).
In fact, you read these measurements the same way you read measurements of an amplifier. In a noise floor measurement, lower is always better. A good converter will sit at -130 dB or lower for a noise floor.
Now it is particularly important to note here that the FFT settings can play a role here! The higher the FFT value, the lower the noise floor can appear…. We test at 32K FFT by default, but will sometimes go a little higher to see if we can look deeper. This may affect the measurements. However, we will always mention it if we deviate from the standard scripts.
You can see that the Sonnet is at -150dB noise floor and the MSB is even 10dB lower. Those are impressive performances. We are guessing that the MSB has shown the limit of the Prism.
Then the frequency response. Here you can see how neutral a converter reproduces (at least: you can see if there are any deviations in the reproduction) as well as what the range is. The range depends on two things: the sampling rate and the filter in use.
Filters
Converters must use filters that limit the bandwidth in relation to the sampling rate. For 16 bit / 44.1 kHz signals, the bandwidth is limited to 22 kHz. With 16 bit / 48 kHz it is 24 kHz and with 24 bit / 96 kHz, for example, it is approximately 48 kHz.
Note: The rule of thumb is that you halve the ‘sampling rate’, although this is not quite correct, since digital filters thus also play a role. For more information, I refer you to Nyquist’s rules .
Now we have already had the digital filter come along a few times. Filters – actually “anti-imaging filters” or also called “reconstruction filters” are necessary in digital audio, because otherwise there would be numerous reflections in the frequency spectrum. Reflections that are going to mix with the analog signal the player sends out to your amplifier: not desirable.
Ideally, we would use a brickwall filter (a very steep filter), but in many cases this gives gigantic phase problems (the steeper a filter, the heavier the phase shift). This is why less steep filters are sometimes used – slow rolloff, for example – or so-called “phase linear” filters that are phase correct (but require more math).
In many cases, the digital filter is a manufacturer’s secret. This is understandable, because it is very sound-determining. And thus determining the quality of the digital player.
We also measure the filters during our tests with a so-called imaging test (See picture above; that’s a typical brickwall filter). This imaging test shows how the filter runs, which says a lot about what philosophy the manufacturer is pursuing.
(See examples of the various filters on the Audiolab Mdac below. We see both brickwall and phase linear as well as some other variations. Which one is better is mostly taste.)
Since June 2023, we also measure the phase response of equipment. Less phase shift means that the phase characteristic of the device is better. This usually manifests itself in a slightly more correct – so not necessarily larger! – stereo image. Below you can see two measurements of phase with the MSB Premier: one in normal mode and one in Theatre Mode. You can immediately see the impact of the filter.
Impulse behavior
The filter also directly affects the impulse behavior of the dac. The impulse behavior shows how a converter reproduces a tight pulse. Ideally, the signal should go straight up and straight down again. But that’s almost impossible.
These measurements on the Audiolob Mdac show beautifully how a filter affects impulse behavior. The “optimal transient” impulse measurement shows almost perfect impulse behavior; almost NOS-like, shall we say. Sharp rolloff almost always gives considerable pre- and post ringing. Slow rolloff already reduces that drastically. Phase linear in this case gives no pre-ringing, but some post-ringing. This is much less annoying than pre-ringing because it is unnatural (in real life we cannot hear something before it is there ;-)).
Linearity
A measurement that also requires some explanation is linearity. This measurement shows whether the input and output signal are, and remain, in proportion. Ideally, this is one diagonal line. And with a decent product this is fortunately the case.
What this measurement also shows is the noise floor of the device. You can see with the Sonnet that – according to the Prism dScope – the Pasithea starts working around -135dB. That corresponds to about 23 bit resolution (24 bit is -144 dB). By now we know that this is the lower limit for the Prism: it cannot measure deeper in this test. :-). Only the Pasithea and MSB achieved this level. Most, normal, converters just reach -120 to about -125 dB is our experience. An exception is the RME ADI which scores very well, but is a tad loud. The Audiolab also does well, the IFI iDSD does not reach -120 dB. And the AudioGD shows a very crazy curve, as you can see below. The result: coloration, as the input and output are out of proportion to each other.
Streamers: sometimes tricky
Now that we’ve given some insight into measuring dacs, let’s move on to streamers. Basically, we measure those in a similar way. After all, if a streamer has a digital input and analog output, we can do the same tests.
Unfortunately, not all streamers are easy to measure. Especially streamers that only have a LAN input are in fact impossible to measure, because we need an input as well as an output; otherwise how do we get a measurement signal inside?
Also a streamer with only digital inputs and outputs is less convenient, because there is no conversion and timing within the digital domain. Therefore, jitter plays no real role, because there is no conversion to analog.
Now we can still measure other things. Think of jitter (on the clock itself, if attainable and possible), since this does play a role when sending a signal to an external dac, and power supply noise via the LISN and noise at the Ethernet port. These are certainly things that say something about the build quality and performance.
You see above, for example, three measurements with the Wavecrest. The idea of these measurements was to show what a power supply does with clock jitter – quite a lot: the jitter halves! You can also see a measurement at the Ethernet port. One measurement with a standard adapter (yellow) and one measurement with a decent power supply (purple). You can see that the noise goes down quite a bit.
So while these are indirect measurements, they actually say just as much. After all, if the power supply is not good, it affects jitter. If there is a lot of internal jitter, a streamer cannot perform decently.
Concluding…
Measuring d/a converters and streamers is fairly automated within Alpha Audio. This is partly due to the scripts in the Prism dScope. The most interesting measurements for d/a converters are – we think – linearity, noise floor and imaging. Perhaps along with dynamic range. This is because it basically tells how the filtering works internally ánd how many bits of resolution the dac achieves in practice.
The same tests apply to streamers. However, these are often not so easy to measure. We measure mainly the noise through the LISN and internally the clock (also for dacs by the way). This also gives a good idea of how well things are arranged. After all: the basis of a device is the power supply. And in the case of digital equipment: also the internal clock.