Chord DAVE Measurements (With MScaler)

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 for testing.

ChitChat:

The DAVE (Digital to Analog Veritas in Extremis) is Chord’s flagship digital to analog converter, and holds quite a high reputation among many audiophiles.
With its lofty reputation (and lofty price!), I was keen to see how this DAC did objectively, and also to take a peek at some of the behaviours or performance aspects that are often touted as being strengths of the DAVE and chord’s pulse array design in general.

This post will primarily be focusing on the DAVE itself, but the M-Scaler will have its own dedicated writeup.

If you’d like to see my review of the DAVE you can watch it here:

Design:

The primary aspect of the Chord DACs, and in particular the DAVE, which really sets them apart from others is the presence of the massively powerful FPGA. This is primarily to enable the DAC to run extremely high performance oversampling filters, with the DAVE having a 164,000 tap filter. Contrast this to most other DACs, which may have filters of only a few hundred taps. The popular ESS9038 Pro chip for example has only 128 taps.

The purpose and effect of high performance oversampling filters is somewhat beyond the scope of this post, but the TLDR is that a higher performance filter enables the DAC to better adhere to Nyquist Theorem and reconstruct the audio waveform more accurately.

One misconception that is relatively important to discuss in regards to Chord’s DAC design is that they are NOT ‘FPGA DACs’. At least, not in the way that many people understand.
In fact, there isn’t really such a thing as an ‘FPGA DAC’. And calling the DAVE an ‘FPGA DAC’ is in some ways doing Chord a disservice by ignoring their genuinely very clever pulse array converter design.

An FPGA is not intended to provide any sort of ‘clean’ output signal even as a 1-bit pulse density modulated stream, and would not make for an ideal converter. In any situation where a DAC is utilising an FPGA, it is either for digital signal processing, or for control of the actual converter circuitry, NOT for doing the actual conversion/output itself.

In dCS DACs, the FPGA controls the Ring DAC converter.
In Holo DACs, the FPGA controls the R2R Ladders
In the RME ADI-2, the FPGA is used to perform DSP before feeding the AKM chip
In the DAVE, the FPGA is used both to perform high quality upsampling, and to control the pulse array.


Many DACs use FPGAs for all sorts of tasks from USB receivers to DSP processors to controlling the actual converter, but from a technical standpoint the actual converter is the important part as that is what separates the various designs. If you were to class any DAC such as the DAVE which used an FPGA to control the discrete converter as an ‘FPGA DAC’ then you’d need to include practically any R2R DAC and any proprietary delta-sigma design such as dCS, T+A, PS Audio, Holo, Denafrips, Metrum etc.
The FPGA is somewhat inconsequential in relation to the converter itself.

The pulse array has a few potential advantages compared to other DAC designs.
The two biggest advantages being that it is extremely resistant to jitter and it is extremely effective at combating noise floor modulation. Something that many delta sigma DACs struggle with. (This is something that ESS has been doing a lot of development work on lately and there are some presentations available on the subject if you’re looking for more information).

Chord DAVE internal, Photo Credit: https://hifistatement.net

The DAVE then has a hybrid discrete output stage, utilising opamps but with the OP portion replaced with a discrete solution.

This feeds the RCA outputs directly, however if using the XLR outputs there is an additional inverting opamp to provide the negative polarity signal.
The DAVE is not a balanced DAC by design, but you do still get the benefit of noise rejection on the interconnects by using the XLR outs because of the additional inverting opamp.

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 (and M-Scaler where mentioned)
– Measurement setup and device under test are running on an AudioQuest Niagara 5000 power supply
– Audioquest Mackenzie XLR and RCA interconnects
– Wave High Fidelity BNC Interconnects
– Intona Reference Impedance Characterized USB Cable
– Exact analyzer/filter configurations for each measurement are detailed in the full reports

Full Measurement Reports:

Reports available here:

Dynamic Range (AES17): 117.6dB

SNR: 119.0dB

Noise Level RMS (20-20khz): 6.217uVrms

Noise Level RMS (20-90khz): 15.27uVrms

DC Offset: 3.585mV active, 1.279mV idle

Latency: 104ms standard, 0.8s with MScaler

1khz 0dBFS Sine, DAC Mode Balanced Out:

In ‘DAC Mode’ the DAC does sometimes show some very low level clipping at full scale as shown above. This seems to be occurring in the digital domain however as when looking at a wider bandwidth we see that the distortion components stop immediately after 22khz. Sometimes this did not happen, sometimes it did, I’m not sure why.

Reducing digital input level by 1dB fixes this:

1khz -1dBFS Sine, DAC Mode XLR Out:

1khz -1dBFS Sine, DAC Mode RCA Out:

The DAVE does actually perform slightly better via the RCA outputs than the XLR outputs at least in terms of THD+N. However when in use with a fully balanced amp or when using longer runs, you’ll probably want to use the XLR outs as the performance penalty on your amp of not doing so will likely be greater than the difference in performance between the DAVE’s RCA and XLR outs.

The DAC can also be put into ‘pre mode’ to enable digital volume control. Doing so introduces no performance penalty. But additionally, turning up the volume does not provide any additional THD+N performance as shown below.

1khz 0dBFS Sine, Pre Mode ‘0dB’ XLR Out:

With a full scale signal, the DAC will start to clip at +3dB, I suggest you keep the DAVE to +2dB or below:

1khz 0dBFS Sine, Pre Mode +3dB XLR Out:

1khz -0.5dBFS Sine, DSD Plus Mode RCA Out:

DSD Performance seems to be essentially identical to PCM, it’s unclear if the DAVE is converting to PCM internally or converting natively.

The DAVE also has an inbuilt headphone amplifier:

1khz Headphone Output, 32 Ohm load, 4v out:

1khz Headphone Output, 32 Ohm load, 700mv out (Headphone level):

1khz Headphone Output, 32 Ohm load, 50mv out (IEM level):

The headphone out also has rising distortion into the higher frequencies:

DAVE headphone output THD+N vs frequency

The headphone output is not bad but not really that great either. It’d definitely be worth getting a higher performance separate amp if you’re considering a DAVE.

HF Filter setting:

The DAVE has a high-frequency filter setting which applies a steep filter to the DAC. This does actually slightly reduce some upper treble content when enabled though not likely to be ‘obvious’ in any case.

Audible band frequency response with HF filter on
HF filter on
HF filter off

-90.31dBFS 1khz sine (96khz capture bandwidth):

The DAVE’s dynamic range even within the 20hz-20khz bandwidth falls behind many other options available today and so lower level signals are not as clean as those produced by either other delta sigma options such as a Gustard X26 Pro or R2R options such as a Holo May.
118dB Dynamic range is still good but given the cost of this device it would have been nice to see some improved performance in this area.

Filter Ultrasonic Attenuation:

THIS is the most interesting part of this DAC, both because there are no other DACs I’m aware of with as intensive/steep a filter as this, but also because somewhat ironically the DAVE’s filter doesn’t actually attenuate that far before the nyquist frequency.

Chord DAVE filter attenuation

The attenuation is actually only at -100dB (roughly 40dB below the white noise spectra) at the nyquist frequency of 22.05khz.
It could certainly be argued that this is still much better than a slower filter that rolls off sooner or another filter that has higher level content further out past 22.05khz, but it’s still odd that Chord didn’t choose to just move things down by less than 0.1khz to ensure that all aliasing is eliminated.
Adding the M-Scaler does not actually change the attenuation by Nyquist, but does drastically steepen the filter:

Chord DAVE + MScaler filter attenuation

Curious that Chord didn’t move the filter just a hair so that full attenuation was achieved by 22.05khz and they could conclusively say it would eliminate aliasing, but still, comfortably the highest performance filter I’ve seen built into any DAC.

Idle Noise FFT:

THD+N vs Frequency:

THD+N vs Frequency, 20khz AES filter

Slight rise in THD+N into the lower frequencies.

IMD:

Left: -108dB, Right: -107dB

Linearity:

With Bandpass Filter
Without Bandpass Filter

Crosstalk:

Jitter:

44.1khz USB
44.1khz via MScaler
48khz USB
48khz via MScaler

Jitter performance is truly excellent both via USB and via the MScaler. However, I would also like to point out that whilst in various bits of marketing Chord has claimed their DACs are ‘immune’ to jitter, this is not the case. They are certainly incredibly RESISTANT to jitter, but not immune. By simulating a bad source and adding some jitter to the digital connection feeding the DAVE we can show jitter performance worsening:

25ns of 1khz sine jitter applied to SPDIF input

So, not quite accurate marketing, but certainly excellent performance.

Multitone:

Wideband:

Idle Noise
1khz tone

Some evidence of switching noise out past 100khz, but quite low level. It’d be interesting to see how this changes with something like a DC4 power supply!

4 thoughts on “Chord DAVE Measurements (With MScaler)”

  1. Sinad by today’s totl measuring dac’s standard is low. The longer high tap filter helps in sharp brick walling to achieve that sharp transient response. I wonder why the harmonics are present reducing the sinad despite so many taps and steep brick walling.

    Reply
    • What sort of measurement error do you feel this could be?
      I do not feel this is the case as there isn’t really anything that’d explain it and the fact that the odd content terminates at the nyquist frequency of the DAC and does not continue for the full bandwidth the analyzer was running at, suggests it is indeed a digital domain issue within the DAC.

      If you have a genuine suggestion I’d be happy to test/address it. But if you’re suggesting it’s measurement error simply because Amir said so (and didn’t provide any reasoning) that’s not particularly constructive

      Reply

Leave a Reply

%d bloggers like this: