Meitner MA3 DAC Measurements

Content and analysis made possible thanks to support from https://headphones.com, and https://patreon.com/goldensound supporters.

This unit was loaned to me by a friend to test

Design:

The Meitner MA3 is a high end digital to analog converter from Canadian manufacturer Meitner, that is a little different from most DACs you might see on the market in that it is a 1-bit or DSD DAC. With a few other unique and/or proprietary design aspects.

The MA3 has a physically very large chassis, though internally is a little bit more empty than you may expect. This isn’t actually an ‘issue’ given as it is a 1-bit DAC which can be relatively simpler in hardware than say an R2R DAC, but I do wish that Meitner had opted for a slightly more compact chassis that would open up this product for desk use in high end headphone setups.

Inside, the blue board toward the top is the digital input board, featuring an XMOS chip for USB input, with some optocoupler ICs for isolation, helping to prevent noise from the connected PC or source affecting the performance of the DAC itself, which in my testing seemed to work very well. Even when cranking prime95 and furmark simultaneously on my Intel i9 + Nvidia 2080TI desktop PC, the DAC exhibited no additional noise through the output nor any degradation in jitter performance.

There is also a small daughterboard which provides network streaming functionality, allowing you to stream directly to the DAC via Roon.

The blue square board toward the centre is the main DSP board and provides the processing power for the MA3’s conversion of all incoming PCM and DSD data to DSD1024. I was a little surprised to see this particular FPGA in use, as it’s about half as powerful as the one in the similarly priced Chord DAVE for instance, and DSD modulators can be quite compute intensive. It would have been nice given the price to perhaps see a little more compute horsepower inside this DAC, similar to what you can find in the Chord DAVE or Playback MPD-8 (though the latter is considerably more expensive than the MA3.)

The actual converters themselves are under the silver IHS units here. Unfortunately I was unable to find images of this converter, and can’t remove the IHS without risking damaging something, but I was able to find an image of the converter from the MA1, and so I would assume that the overall design will be similar to that of the MA3 (though as said I cannot be certain).

This seems to be a moving average FIR filter approach, similar to that employed by the Holo May or T+A DAC200. Rather than having one single element switching at the full DSD rate, from 100% to 0% each time a sample is different from the previous one, you instead have a shift register, into which you load say 5 samples, and the output averages the value of each. This means that each time a new sample arrives, the maximum value change at the output would be 20% not 100%, reducing switching noise quite drastically. This is an effective method of both alleviating some of the challenges of designing a ‘pure’ 1-bit switching circuit, whilst filtering out much of the resulting switching noise created by the DSD format inherently, without needing to put things through a capacitor or other analog low-pass filter circuit first.

Measurements:

Test Setup:

– Audio Precision APx555 B-Series Analyzer with 200kOhm input impedance set unless otherwise specified
– USB Source: Intel PC via intona 7055-C isolator
– All measurements shown are with the DAC connected via USB unless otherwise specified
– Measurement setup and device under test are running on regulated 230V power from a Furman SPR-16-Ei and an AudioQuest Niagara 3000
– Audioquest Mackenzie XLR and RCA interconnects
– Intona Reference Impedance Characterized USB Cable
– Audioquest Carbon SPDIF and AES cables (1.5m)
– Exact analyzer/filter configurations for each measurement are detailed in the full reports
– Measurements shown are with the DAC set to no attenuation unless otherwise stated

Full Measurement Reports:

Reports available here:

Dynamic Range (AES17): 111.4dB

SNR: 113.1dB

IMD SMPTE: -94.6dB

DC Offset: 1.0mV active, 1.2mV idle

Output Impedance: 299 Ohms (XLR and RCA)

1khz Sine (PCM Input), 0dBfs, XLR Out:

1khz Sine (PCM Input), 0dBfs, RCA Out:

RCA and XLR performance is identical, and output impedance is the same for both (though voltage is 2x on XLR). It’s unclear whether this DAC is truly balanced or not.

1khz Sine (PCM Input), -3dBfs, XLR Out:

1khz Sine (DSD Input), -3dBfs (HQPlayer vol), XLR Out:

I wanted to check whether feeding the MA3 DSD would provide a benefit in performance when using the high performance modulators from HQPlayer. Unfortunately, it seems that the MA3 will actually put DSD info through its own processing anyway, evidenced by the fact that the digital volume still works, and the ultrasonic noise profile as shown below, still shows the same ultrasonic noise profile from its own modulator, just with whatever was contained in the input DSD info also added on. The MA3 only accepts DSD128 input, which means that you’re going to have added noise from HQP’s modulator ontop of the DAC’s own. Whereas you likely wouldn’t if it accepted DSD256/512. Therefore, I would recommend feeding the MA3 normal PCM, though external upsampling does help to overcome the DACs quite slow filter design.

Jitter:

Jitter performance is overall very good, some low frequency jitter/phase noise visible but not much.

Upsampling Filter:

The MA3 uses a very slow upsampling filter, which means at 20khz content is attenuated by about 4dB, whilst failing to attenuate unwanted content above the Nyquist frequency effectively at all. This is unfortunate, both because having a faster filter would in my view be desirable, but also because it DOES actually have a filter that is faster with a flatter frequency response, you just cannot choose to use it!

The MA3 has a system that looks at the content of the incoming data, and selects a different filter depending on whether there is transient content present. Playing a single sine through the DAC rather than more complex white noise shows a much flatter frequency response than what we saw above, being only 0.3dB down at 20khz, not 4dB:

Static sines seem to cause the DAC to use its faster filter, and playing anything exclusively below 1.5khz does so too, but when the DAC detects the presence of more ‘complex’ content above 1.5khz, it switches to the slower filter. We can demonstrate this by playing an 18khz sine with a 1khz sine, and then swapping it to 1.6khz. As you can see, once you add the 1.6khz sine in, it suddenly switches and now the 18khz sine gets attenuated by the slower filter’s treble rolloff, and there is quite a bit of aliasing visible above 20khz.

Additionally, if that 2nd sine is set to 1.501khz, the DAC seems to get confused and just keep switching back and forth between filters.

The intent seemed to be that the DAC would respond in real time to transients in the music, however in experimenting, I found that the DAC will always use the slower filter so long as there is complex content present above 1.5khz and with an amplitude of at least -80dBfs, which means basically everything. I also tried conducting a test where I played a track with a single sine, then suddenly added a second, and it took the DAC around 4-5ms to respond. 1khz sound has a period of 1ms, and 20khz is 0.05ms, so the DAC is not responding quickly enough to react to individual transients, but rather just the general content makeup over a longer period of time. I really wish that there was an option to disable this automatic switching entirely, and just let the user choose whether to use the slower or faster filter.

Low level signal output:

Low level signal accuracy is a bit of a mixed bag here. The DAC produces a moderate amount of ultrasonic noise which obscures the signal quite a bit. If we set the analyzer to filter that out and just leave the content under 20khz, then things look better (linearity is still ‘just ok’ though compared to competitors, as shown later down in this post). So performance in the audible band is ok, it’s just that small signals are somewhat drowned out by out-of-band noise.

Idle Noise FFT:

There is a moderate amount of ultrasonic noise due to the nature of it being a 1-bit DAC. It would have been nice to see either a slightly better modulator or more effective analog filtering in order to reduce this given the hefty price of this unit. The Holo May for example when fed DSD from HQPlayer and converting on its own moving-average FIR DSD converter, has drastically less remaining noise.

THD+N vs Frequency:

Flat as a pancake! Good stuff.

IMD:

IMD performance is good.

Linearity:

There is a sudden increase in THD at about -25dBfs of nearly 20dB. This means that higher level signals distort comparatively less vs lower level signals.

Crosstalk:

Multitone:

Intersample Overs Test:

When DACs oversample, they can sometimes encounter a situation where the reconstructed/interpolated waveform goes above 0dBfs (the maximum possible digital value).
This is particularly common with poorly mastered music that has been ‘brickwalled’ (To clarify: I mean brick-walled as in the loudness-wars term, not brick-wall nyquist reconstruction filters https://en.wikipedia.org/wiki/Loudness_war ).

DACs ideally should have a few dB of digital headroom to accommodate this and reconstruct properly, and many such as Chord, RME, Benchmark etc do, and will output signals with intersample overs without any issue/distortion.
But many do not do this, either out of lack of awareness of the issue or because doing so usually means sacrificing a few dB of dynamic range and/or THD+N performance. And as a result will not be able to properly reconstruct these signals.

We can test this by creating a signal that induces intersample overs.
See in the image below. All the 44.1khz samples (squares) are below the 0dBfs limit, but the actual waveform itself, is above this limit.

The MA3 perfectly handles intersample overs and will not clip if they are present in the source material.

Summary:

The Meitner MA3 offers decent but not excellent performance, and has a few quirks that make it a little tough to recommend based on the objective performance. This is an extremely expensive product, but the poor small-signal accuracy and slow rolloff filter alone seem to (in my subjective opinion) provide an overall softer sound that leaves it lacking in detail and dynamics compared to various products at and below its pricepoint.

The technology is nice, and with the R&D required to make a 1-bit DAC from scratch, the price is not unjustified, it just unfortunately does not hold up objectively in most areas to cheaper options.

The video review with my full subjective opinions will be up shortly!