DIYINHK DAC Measurements

[mdsimon2]

A few times on this board and specifically recently in the ASR open source streamer thread I have seen folks express interest in the DIYINHK multichannel DACs based on the ES9016 and ES9038pro chips. From reading about these DACs on DIY Audio I have seen two groups of comments, 1) implementation is not great (especially the early 2 layer boards) and 2) some folks have built the ES9016 and were quite happy with it. Thought I would try to bridge the two by making some measurements and listening. I have some basic test equipment (scope, Motu M4, DMM) and have been gaining some experience with making measurements. More than happy to experiment so if there is something you want to see feel free to ask.

Goals
-Learn how power supplies impact performance
-Build low cost platform to learn CamillaDSP, specifically with DIY active speakers in mind
-Learn how different op amps impact performance
-Build low cost but decent performance miniSHARC based DSP (like the discontinued OpenDRC-DA8 with a better DAC)
-Learn DAC register programming with an Arduino

ES9016
-8 channel, I2S input DAC
-Single ended output with TI NE5532 op amps, presumably based on Figure 3 of ESS Application Note Maximizing DAC Performance for Every Budget
-Comes with no power supply and has no onboard regulators, requires regulated 2 x 3.3V and +/- 12V
-In typical operation DAC is clocked synchronously from I2S source but board has pads to allow installation of a local oscillator to allow for asynchronous clocking

Some of this stuff I already had on hand but see below for a BOM with prices to get this thing working. Might be able to get a cheaper transformers / regulators but imagine most builds are in the $250-300 range.

USB DAC - $285 TOTAL
ES9016 DAC - $90 - DIYINHK
multichannel XMOS - $70 - DIYINHK
3.3V x 2 regulator - $40 - DIYINHK
+/- 12V regulator - $40 - DIYINHK
6V x 2 transformer - $20 - Mouser / DigiKey
15V x 2 transformer - $20 - Mouser / DigiKey
fused IEC receptacle / switch - $3 - Parts Express
XH cables (2 x red/black, 1 x yellow/black/blue) - $2 - DIYINHK

Here is a pic of the DAC assembly.

This is pretty cost competitive with the now discontinued miniDSP U-DAC8 ($255) and hopefully offers improved performance. Assembly is very easy if you have done any thru hole soldering, takes about 30-60 minutes to complete. I am not a huge fan of the DIP8 sockets provided so I used some slightly more robust sockets which adds $11 to the build.

All measurements below were done with a MOTU M4 audio interface. I did not find any differences across channels. Playback and measurement sample rates were the same. 128K FFT length, Blackman-Harris 7 windowing and 8 averages were used for all measurements.

1 kHz THD+N, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 1/2.

Channel 1
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.83 dBFS
-10.8 dBFS C, -10.8 dBFS A
-10.8 dBFS 22 - 22k UNW
Distortion at 999.9 Hz, -10.8 dBFS:
THD: -108.3 dB based on 21 harmonics [20..22000 Hz]
HHD: -122.2 dB [10 .. 22]
N: -99.8 dB [20..22000 Hz]
N+D: -111.2 dBFS A
THD+N: -99.2 dB [20..22000 Hz]
2nd harmonic -115.2 dB
3rd harmonic -139.4 dB
4th harmonic -113.4 dB
5th harmonic -120.7 dB
6th harmonic -115.5 dB
7th harmonic -118.7 dB
8th harmonic -122.0 dB
9th harmonic -120.3 dB

Channel 2
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.82 dBFS
-10.8 dBFS C, -10.8 dBFS A
-10.8 dBFS 22 - 22k UNW
Distortion at 999.9 Hz, -10.8 dBFS:
THD: -109.0 dB based on 21 harmonics [20..22000 Hz]
HHD: -122.5 dB [10 .. 22]
N: -99.6 dB [20..22000 Hz]
N+D: -111.1 dBFS A
THD+N: -99.1 dB [20..22000 Hz]
2nd harmonic -122.1 dB
3rd harmonic -139.6 dB
4th harmonic -113.4 dB
5th harmonic -120.2 dB
6th harmonic -115.6 dB
7th harmonic -118.2 dB
8th harmonic -121.4 dB
9th harmonic -120.6 dB

THD is quite low at -108 dB. The noise from this measurement is definitely impacted by the MOTU M4 noise. For example see measurement below from an Okto dac8 pro at a similar level (within 1 dB), here the MOTU M4 noise dominates the measurement limiting THD+N to a similar -99 dB.

Frequency response, 44.1 kHz sample rate. Channel 1/2.

No surprises here, response is flat to 20 kHz.

THD and THD+N vs frequency, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 1.

THD+N (upper trace) and THD (lower trace) are well behaved with little variation with frequency.

THD vs level, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 1.

Bit of an "ESS Hump" going on here.

J Test, 44.1 kHz sample rate. Channel 1/2.

Pretty clean, maybe some close-in spikes at -120 dB.

AES-17 MD, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 1/2.

Channel 1
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -12.82 dBFS
-14.8 dBFS C, -26.3 dBFS A
-12.8 dBFS 22 - 22k UNW
IMD is -106.3 dB
f1 = 41 Hz, f2 = 7,993 Hz
IMD components:
d2L: -118.0 dB
d2H: -106.6 dB
d3L: -114.7 dB
d3H: -114.0 dB
d4L: -103.3 dB
d4H: -103.1 dB
d5L: -106.5 dB
d5H: -106.6 dB
TD+N is -97.6 dB

Channel 2
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -12.82 dBFS
-14.8 dBFS C, -26.3 dBFS A
-12.8 dBFS 22 - 22k UNW
IMD is -105.9 dB
f1 = 41 Hz, f2 = 7,993 Hz
IMD components:
d2L: -126.7 dB
d2H: -105.9 dB
d3L: -113.7 dB
d3H: -112.3 dB
d4L: -103.4 dB
d4H: -103.4 dB
d5L: -106.1 dB
d5H: -106.9 dB
TD+N is -97.5 dB

AES-17 DFD, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 1/2.

Channel 1
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -14.96 dBFS
-25.5 dBFS C, -23.6 dBFS A
-15.0 dBFS 22 - 22k UNW
IMD is -97.2 dB
f1 = 18,000 Hz, f2 = 20,000 Hz
IMD components:
d2L: -111.2 dB
d2H: N/A
d3L: -108.2 dB
d3H: -106.9 dB
d4L: -126.8 dB
d4H: N/A
d5L: -142.6 dB
d5H: N/A
TD+N is -95.9 dB

Channel 2
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -14.96 dBFS
-25.5 dBFS C, -23.6 dBFS A
-15.0 dBFS 22 - 22k UNW
IMD is -96.5 dB
f1 = 18,000 Hz, f2 = 20,000 Hz
IMD components:
d2L: -110.1 dB
d2H: N/A
d3L: -107.5 dB
d3H: -106.3 dB
d4L: -126.5 dB
d4H: N/A
d5L: -144.0 dB
d5H: N/A
TD+N is -95.6 dB

Next set of measurements is the same as above but at 96 kHz sample rate. In general behavior is the same, some of the THD+N values are worse as these measurements use a 20-48000 Hz bandwidth which incorporates more noise (in addition the ADC has rising noise at higher frequencies), if limited to 20-22000 kHz these would be the same as the 44.1 kHz sample rate values.

1 kHz THD+N, 96 kHz sample rate, 20-48000 Hz bandwidth. Channel 1/2.

Channel 1
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.79 dBFS
-10.8 dBFS C, -10.8 dBFS A
-10.8 dBFS 22 - 22k UNW
-109.7 dBFS >22k
Distortion at 999.8 Hz, -10.8 dBFS:
THD: -107.4 dB based on 47 harmonics [20..48000 Hz]
HHD: -119.4 dB [10 .. 48]
N: -96.3 dB [20..48000 Hz]
N+D: -110.8 dBFS A
THD+N: -96.0 dB [20..48000 Hz]
2nd harmonic -113.6 dB
3rd harmonic -126.0 dB
4th harmonic -113.8 dB
5th harmonic -118.7 dB
6th harmonic -115.5 dB
7th harmonic -116.6 dB
8th harmonic -120.4 dB
9th harmonic -119.2 dB

Channel 2
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.79 dBFS
-10.8 dBFS C, -10.8 dBFS A
-10.8 dBFS 22 - 22k UNW
-109.7 dBFS >22k
Distortion at 999.8 Hz, -10.8 dBFS:
THD: -108.2 dB based on 47 harmonics [20..48000 Hz]
HHD: -119.6 dB [10 .. 48]
N: -96.3 dB [20..48000 Hz]
N+D: -111.1 dBFS A
THD+N: -96.1 dB [20..48000 Hz]
2nd harmonic -119.1 dB
3rd harmonic -127.1 dB
4th harmonic -113.9 dB
5th harmonic -118.4 dB
6th harmonic -115.5 dB
7th harmonic -116.4 dB
8th harmonic -120.6 dB
9th harmonic -119.2 dB

Frequency response, 96 kHz sample rate. Channel 1/2.

THD vs frequency, 96 kHz sample rate, 20-48000 Hz bandwidth. Channel 1.

THD vs level, 96 kHz sample rate, 20-48000 Hz bandwidth. Channel 1.

J Test, 96 kHz sample rate. Channel 1/2.

AES-17 MD, 96 kHz sample rate, 20-48000 Hz bandwidth. Channel 1/2.

Channel 1
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -12.82 dBFS
-14.8 dBFS C, -26.3 dBFS A
-12.8 dBFS 22 - 22k UNW
-109.5 dBFS >22k
IMD is -105.9 dB
f1 = 41 Hz, f2 = 7,993 Hz
IMD components:
d2L: -119.3 dB
d2H: -106.1 dB
d3L: -109.7 dB
d3H: -110.3 dB
d4L: -104.1 dB
d4H: -104.2 dB
d5L: -104.4 dB
d5H: -105.1 dB
TD+N is -94.5 dB

Channel 2
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -12.81 dBFS
-14.8 dBFS C, -26.3 dBFS A
-12.8 dBFS 22 - 22k UNW
-109.5 dBFS >22k
IMD is -105.1 dB
f1 = 41 Hz, f2 = 7,993 Hz
IMD components:
d2L: -123.4 dB
d2H: -105.2 dB
d3L: -110.6 dB
d3H: -112.3 dB
d4L: -104.6 dB
d4H: -104.2 dB
d5L: -104.0 dB
d5H: -104.7 dB
TD+N is -94.5 dB

AES-17 DFD, 96 kHz sample rate, 20-48000 Hz bandwidth. Channel 1/2.

Channel 1
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -14.93 dBFS
-25.5 dBFS C, -23.6 dBFS A
-14.9 dBFS 22 - 22k UNW
-108.8 dBFS >22k
IMD is -94.5 dB
f1 = 18,000 Hz, f2 = 20,000 Hz
IMD components:
d2L: -110.2 dB
d2H: -109.1 dB
d3L: -107.6 dB
d3H: -102.4 dB
d4L: -126.5 dB
d4H: -116.5 dB
d5L: -137.5 dB
d5H: -121.9 dB
TD+N is -91.3 dB

Channel 2
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -14.93 dBFS
-25.5 dBFS C, -23.6 dBFS A
-14.9 dBFS 22 - 22k UNW
-108.8 dBFS >22k
IMD is -94.5 dB
f1 = 18,000 Hz, f2 = 20,000 Hz
IMD components:
d2L: -108.9 dB
d2H: -109.6 dB
d3L: -107.8 dB
d3H: -102.6 dB
d4L: -125.7 dB
d4H: -115.8 dB
d5L: -140.0 dB
d5H: -122.2 dB
TD+N is -91.4 dB

Overall measured performance is quite acceptable. I also tried OPA1612 op amps but was unable to measure a performance improvement although it is possible that this is limited by the noise of my measurement gear. Overall see no reason to change op amps from the base TI NE5532 op amps.

1 kHz THD+N, 44.1 kHz sample rate, 20-22000 Hz. Channel 1, OPA1612 op amp.

Channel 1
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.81 dBFS
-10.8 dBFS C, -10.8 dBFS A
-10.8 dBFS 22 - 22k UNW
-109.7 dBFS >22k
Distortion at 1,000.0 Hz, -10.8 dBFS:
THD: -108.7 dB based on 21 harmonics [20..22000 Hz]
HHD: -120.3 dB [10 .. 22]
N: -100.0 dB [20..22000 Hz]
N+D: -111.5 dBFS A
THD+N: -99.5 dB [20..22000 Hz]
2nd harmonic -127.7 dB
3rd harmonic -136.3 dB
4th harmonic -114.0 dB
5th harmonic -119.3 dB
6th harmonic -115.1 dB
7th harmonic -116.9 dB
8th harmonic -120.7 dB
9th harmonic -118.9 dB

I tried using a SilentSwitcher power supply with a 9V / 2A MeanWell SMPS as an alternative power supply as it is much smaller (although similar in cost to the linear power supply). Performance is significantly degraded from noise although the 3.3V supply is not exceptionally low noise so this was expected.

1 kHz THD+N, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 1, SilentSwitcher power supply.

Channel 1
131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.63 dBFS
-10.6 dBFS C, -10.6 dBFS A
-10.6 dBFS 22 - 22k UNW
Distortion at 1,000.0 Hz, -10.6 dBFS:
THD: -90.3 dB based on 21 harmonics [20..22000 Hz]
HHD: -97.0 dB [10 .. 22]
N: -82.6 dB [20..22000 Hz]
N+D: -92.9 dBFS A
THD+N: -81.9 dB [20..22000 Hz]
2nd harmonic -97.0 dB
3rd harmonic -113.2 dB
4th harmonic -110.6 dB
5th harmonic -116.8 dB
6th harmonic -115.8 dB
7th harmonic -95.9 dB
8th harmonic -120.3 dB
9th harmonic -95.9 dB

Adding 3.3V LDOs improves performance but still not quite as good as linear power supply, still a respectable option if you are trying to save space.

1 kHz THD+N, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 1, SilentSwitcher power supply with 3.3V LDOs.

131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.81 dBFS
-10.8 dBFS C, -10.8 dBFS A
-10.8 dBFS 22 - 22k UNW
-104.0 dBFS >22k
Distortion at 999.8 Hz, -10.8 dBFS:
THD: -102.1 dB based on 21 harmonics [20..22000 Hz]
HHD: -116.9 dB [10 .. 22]
N: -95.8 dB [20..22000 Hz]
N+D: -108.4 dBFS A
THD+N: -94.9 dB [20..22000 Hz]
2nd harmonic -104.2 dB
3rd harmonic -123.5 dB
4th harmonic -114.2 dB
5th harmonic -118.8 dB
6th harmonic -115.4 dB
7th harmonic -111.4 dB
8th harmonic -120.2 dB
9th harmonic -112.8 dB

Listening
Turn on/off pop is pretty bad, definitely recommend proper power sequencing (turning on DAC followed by power amplifiers, turning off power amplifiers followed by DAC). No hiss or noise noted with normal sensitivity speakers (85-90 dB). Sounds excellent! I personally do not much attribute much sonic performance to DACs but overall the DAC sounds good. For now my listening is confined to 2 channel passive speakers as I am still getting CamillaDSP up and running. Tried both a MacBook Pro and Raspberry Pi4 as sources, both were good. Overall for a sub-$300 8 channel USB DAC I think this does quite well. I do not have a U-DAC8 but I do have an OpenDRC-DA8 which is based on the same underlying DAC board and the biggest difference is the substantially higher noise (audible hiss) on the OpenDRC-DA8. Excited to get CamillaDSP up and running and trying this out on some DIY active speakers.

Michael

 

[mdsimon2]

Next step was to try and interface with the miniSHARC. I had limited confidence that this would work well in the standard synchronous mode as I have not had much luck in the past clocking DACs / digital output boards directly from the miniSHARC. I first tried using the miniSHARC with a DIGI-FP as this provides 3 digital inputs. I could not get this to successfully work presumably because the DIGI-FP is also loading the master clock output (DIGI-FP is slaved to the miniSHARC). I then tried using a cheap AK4118 SPDIF input board instead of the DIGI-FP and using the miniSHARC's internal ASRC (input slave mode). In this configuration there is no master clock connection from the miniSHARC to AK4118 board so hopefully less master clock loading. This configuration worked well for up to 4 outputs from the miniSHARC but if 6 or 8 channel outputs were connected it would not work. It is possible if I played around with termination resistors this might work but have not attempted this yet. My hope was that by adding a local 100 MHz oscillator and operating in asynchronous mode the miniSHARC would play nice.

For this setup I installed a 30 pin male header with the male pins facing up so that it would be easier to connect the miniSHARC. I used a 100 MHz Crystek 575X oscillator and 0.01 uF bypass capacitor as recommended by the Crystek data sheet. There are certainly cheaper clock options available in 80/100 MHz (Crystek CC3391/2, NDK SD) but figured if I was installing a clock I should use a good low phase noise clock, the 575 package is also much more robust and easier to solder than the very small NDK SD. In this setup the miniSHARC worked perfectly with the DIGI-FP!

Measured performance is identical to the synchronously clocked XMOS. The advantage of this setup is that it can be used as a more traditional DAC, with 3 digital inputs (AES, SPDIF, TOSLINK) and the miniSHARC can be used for volume control. BOM and cost breakdown for this setup are shown below.

miniSHARC DAC - $474
ES9016 DAC - $90 - DIYINHK
3.3V x 2 regulator - $40 - DIYINHK
+/- 12V regulator - $40 - DIYINHK
6V x 2 transformer - $20 - Mouser / DigiKey
15V x 2 transformer - $20 - Mouser / DigiKey
fused IEC receptacle / switch - $3 - Parts Express
XH cables (2 x red/black, 1 x yellow/black/blue) - $2 - DIYINHK
miniSHARC - $185 - miniDSP
DIGI-FP - $45 - miniDSP
Crystek 575X 100 MHz - $27 - Mouser / DigiKey
0.01 uF 0603 bypass capacitor - $2 - Mouser / DigiKey

This is ~$100 more than the discontinued OpenDRC-DA8 ($375) but offers a much higher performance DAC (20+ dB SINAD improvement) and has 3 digital inputs instead of 1. This is also less than a 4X10HD but has a more powerful DSP and a slightly better performing DAC, however the 4X10HD has balanced outputs and analog inputs which this SHARC DAC does not have. I tried this DAC with my normal DIY active speaker setup (LXmini + Dayton RSS315HFA subs) and it works great. Amplifiers are Audiophonics Hypex NC252MPs with XLR inputs (I used mono price RCA to XLR cables).

See below for some pics of the setup with the miniSHARC and the installed oscillator.

In comparison my normal DSP / DAC setup is a DIY miniSHARC with 4x AES output and an Okto DAC8 pro. Other than missing the very nice display, trigger and complete lack of on/off pops of the Okto DAC8 pro this is a worthy alternative at 1/4 of the price.

I did also purchase the 9038pro based DAC but honestly had lower expectations for that DAC as I understand properly implementing a 9038pro is difficult. This DAC has a local oscillator installed and operates asynchronously. Measured performance is worse than the ES9016 and shows different performance depending on channel. Channel 1/2 and 7/8 (edges of DAC board) have a SINAD of 92 dB and Channels 3/4 and 5/6 have a SINAD of 98 dB (middle of DAC board). I also experienced intermittent noise with this DAC that I do not quite understand. With the base XMOS everything seemed fine, however when using the miniSHARC I would often get noticeable noise, especially when using all 8 channels. The weirdest thing is that once this occurred even if I switched back to the XMOS I would still get noise! The only way to solve it was to shut off the DAC for several minutes and start over. It is possible I made a bad solder joint on the board or something else is damaged but I haven't found the culprit yet. To be honest I am not that interested in pursuing the ES9038pro further as the ES9016 has good measured performance and is rock solid. Would definitely be interested if anyone has played around with the 9038pro board and has experience to share.

Hope this was useful for anyone considering these DACs, if you have a questions please let me know. Next step is to get CamillaDSP up and running on the Pi and start playing around with programming the DAC registers.

Michael

 

[restorer-john]

Your impulse response is showing the combination of the D/A and the A/D front end in the Motu. It also suggests to me that the polarity is inverted in your D/A.

 

[mdsimon2]

Your impulse response is showing the combination of the D/A and the A/D front end in the Motu. It also suggests to me that the polarity is inverted in your D/A.

Good call, thanks for the feedback. I mistakenly glanced at the impulse response and thought pre-ringing = linear phase but I agree that something is flipping polarity. I will make some acoustic measurements to see what is causing the inversion (DAC or ADC). In the mean time I deleted the impulse response plots.

Michael

 

[diyinhk]

A user told us about this link, I hope to answer here to avoid the same question asking us again and again

1. All dac chip has pop noise during power on/off, AK449x is the largest I have ever encounter and ES9038 is the lowest, es9016 is larger than es90x8. All components in the audio path can also create power on click noise includes preamp and main amp. To avoid redundant click noise eliminator in the audio path to affect the signal quality, a single click noise eliminator using a mechanical relay before the speaker should be the best choice. click noise eliminator is quite simple, power on delay and instant off can be done by a RC and transistor circuit. Use a power strip to turn on all equipment at once, the mechanical relay will turn on the speaker after 2 second.

2. Connect the multichannel xmos pcb directly to the dac pcb without any cable as the default setup is the way to achieve stable playback and same performance over different channel. Long and loose i2s cable without dedicated ground return degrade the sound quality and unnormal playback is normal.

3. During our test, turning on/off of the soldering iron behind can also interrupt the dac playback. Depends on the room emi condition, dac with metal enclosure can increase the measured performance. Be careful do not connect the dac power ground to the metal enclosure, the RCA socket is connected to the power ground and it should not touch the metal enclosure.

 

[LTig]

I think the OP should show the MOTU loopback plots so it is easier to see the limits of his rig.

And in REW one can show the distortion window within the RTA window so one needs not to write down tve results manually below the plot.

 

[mdsimon2]

I think the OP should show the MOTU loopback plots so it is easier to see the limits of his rig.

And in REW one can show the distortion window within the RTA window so one needs not to write down tve results manually below the plot.

As mentioned above I posted a FFT of an Okto dac8 pro (which is better performing than the MOTU DAC) in to the MOTU ADC at the same output level as the ES9016 which shows that overall performance is limited by the ADC noise. I’ll run the other tests when I have a chance and post those as well.

I am aware of the ability to show the distortion measurements in the RTA window (again see Okto measurement in first post which does this). To my knowledge you can only do this for one channel at a time and it needs to be screenshot at the time of measurement. As a result I prefer to save the measurement in REW which will record the data as shown above in the measurement comments and allow for combining of multiple traces post-measurement, YMMV.

Michael

 

[mdsimon2]

Have a few more measurements to bring to the table. Here is the MOTO M4 loopback. This is an interesting comparison as the MOTU also uses an ES9016 DAC although as it only has 4 channels of output I assume 2 channels are paralleled for each output so not an exact comparison. It is also interesting in that Amir has measured the MOTU so we have an idea of the level of performance of the DAC and ADC separately. Overall performance was very similar to the DIYINHK measurements.

MOTU M4 loopback. 1 kHz THD+N, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 3 output, channel 4 input.

A bit better from a THD perspective but you can see that the overall THD+N is limited by noise, presumably from the ADC.

MOTU M4 loopback. AES-17 MD, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 3 output, channel 4 input.

MOTU M4 loopback. AES-17 DFD, 44.1 kHz sample rate, 20-22000 Hz bandwidth. Channel 3 output, channel 4 input.

MOTU M4 loopback. Frequency response, 44.1 kHz sample rate. Channel 3 output, channel 4 input.

Michael

 

[mdsimon2]

All prior measurements of the DIYINHK DAC were with 1 output channel only. When I started playing around with the miniSHARC setup I noticed that if more channels were output performance degraded as a result of an elevated 2nd harmonic. I had a hunch that this may have been related to the length of wiring I had between the power supplies and DAC. Original wiring was 25 cm long but I also had some pre-crimped 16 cm long wiring on hand. 16 cm long wiring resulted in a -5 dB THD+N improvement with 4 channels playing. I then shortened the wiring even further to just a few cm and THD+N performance improved by another -3 dB. I was also curious if shorter wiring would improve the performance of the SilentSwitcher power supply and sure enough the performance of the SilentSwitcher with 4 channels playing was actually around -1 dB better than the linear power supply. However the SilentSwitcher performance is dominated by noise and the linear power supply is dominated by the 2nd harmonic so the linear power supply may sound better in practice. I would really like to use the SilentSwitcher as it is so much smaller than the linear power supply. Either way in my final implementation I will use as short as possible wiring likely directly soldered from the power supply to the DAC board.

DIYINHK ES9016 25 cm wiring, linear PS. 1 kHz THD+N. 44.1 kHz sample rate, 22-22000 Hz bandwidth. Channel 1, 3, 5, 7 output, channel 1 measured. -88.4 dB THD+N.

131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.42 dBFS
-10.4 dBFS C, -10.4 dBFS A
-10.4 dBFS 22 - 22k UNW
Distortion at 999.9 Hz, -10.4 dBFS:
THD: -88.6 dB based on 21 harmonics [20..22000 Hz]
HHD: -120.0 dB [10 .. 22]
N: -99.9 dB [20..22000 Hz]
N+D: -97.6 dBFS A
THD+N: -88.3 dB [20..22000 Hz]
2nd harmonic -88.6 dB
3rd harmonic -121.8 dB
4th harmonic -113.9 dB
5th harmonic -123.2 dB
6th harmonic -115.9 dB
7th harmonic -117.5 dB
8th harmonic -121.5 dB
9th harmonic -120.7 dB

DIYINHK ES9016 16 cm wiring, linear PS. 1 kHz THD+N. 44.1 kHz sample rate, 22-22000 Hz bandwidth. Channel 1, 3, 5, 7 output, channel 1 measured. -93.4 dB THD+N.

131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.41 dBFS
-10.4 dBFS C, -10.4 dBFS A
-10.4 dBFS 22 - 22k UNW
Distortion at 999.9 Hz, -10.4 dBFS:
THD: -94.5 dB based on 21 harmonics [20..22000 Hz]
HHD: -119.6 dB [10 .. 22]
N: -99.9 dB [20..22000 Hz]
N+D: -103.1 dBFS A
THD+N: -93.4 dB [20..22000 Hz]
2nd harmonic -94.7 dB
3rd harmonic -122.1 dB
4th harmonic -114.0 dB
5th harmonic -124.2 dB
6th harmonic -115.9 dB
7th harmonic -117.3 dB
8th harmonic -121.4 dB
9th harmonic -120.7 dB

DIYINHK ES9016 short wiring, linear PS. 1 kHz THD+N. 44.1 kHz sample rate, 22-22000 Hz bandwidth. Channel 1, 3, 5, 7 output, channel 1 measured. -95.9 dB THD+N.

131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.41 dBFS
-10.4 dBFS C, -10.4 dBFS A
-10.4 dBFS 22 - 22k UNW
Distortion at 999.9 Hz, -10.4 dBFS:
THD: -98.5 dB based on 21 harmonics [20..22000 Hz]
HHD: -119.4 dB [10 .. 22]
N: -99.4 dB [20..22000 Hz]
N+D: -106.1 dBFS A
THD+N: -95.9 dB [20..22000 Hz]
2nd harmonic -98.8 dB
3rd harmonic -123.3 dB
4th harmonic -114.3 dB
5th harmonic -123.5 dB
6th harmonic -115.9 dB
7th harmonic -117.7 dB
8th harmonic -121.3 dB
9th harmonic -121.2 dB

DIYINHK ES9016 short wiring, SS+LDO PS. 1 kHz THD+N. 44.1 kHz sample rate, 22-22000 Hz bandwidth. Channel 1, 3, 5, 7 output, channel 1 measured. -96.7 dB THD+N.

131072-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -10.45 dBFS
-10.5 dBFS C, -10.5 dBFS A
-10.5 dBFS 22 - 22k UNW
Distortion at 999.9 Hz, -10.5 dBFS:
THD: -102.9 dB based on 21 harmonics [20..22000 Hz]
HHD: -115.9 dB [10 .. 22]
N: -97.9 dB [20..22000 Hz]
N+D: -108.3 dBFS A
THD+N: -96.7 dB [20..22000 Hz]
2nd harmonic -104.4 dB
3rd harmonic -117.2 dB
4th harmonic -113.9 dB
5th harmonic -126.1 dB
6th harmonic -115.9 dB
7th harmonic -116.8 dB
8th harmonic -120.9 dB
9th harmonic -121.2 dB

And finally a picture of the SilentSwitcher + LDO PS with the shorter wiring.

Overall this thing has been a lot of fun and I am learning a lot.

Michael

 

[fluid]

There are a couple of things I did different with my build that may or may not have made much difference.
I replaced the bulk capacitors on the board with Nichicon, the main difference being a solid polymer cap on the digital power supply for the very low ESR.

I used Analog Devices ADA4898 SMT opamps directly soldered to the boards with the pads provided mainly due to the noise spec and because I wanted to try them. Whilst the NE5532 opamps are good there are better measuring parts available like the LM4572 which would be an easy swap. With there being eight of them it could perhaps make some difference.

I also got a custom R-Core Transformer from China . They are quieter than toroids from a background hum point of view and a single one took up less space than multiple small toroids if space is a concern.

I still haven't actually finished mine yet.

 

[mdsimon2]

There are a couple of things I did different with my build that may or may not have made much difference.
I replaced the bulk capacitors on the board with Nichicon, the main difference being a solid polymer cap on the digital power supply for the very low ESR.

I used Analog Devices ADA4898 SMT opamps directly soldered to the boards with the pads provided mainly due to the noise spec and because I wanted to try them. Whilst the NE5532 opamps are good there are better measuring parts available like the LM4572 which would be an easy swap. With there being eight of them it could perhaps make some difference.

I also got a custom R-Core Transformer from China . They are quieter than toroids from a background hum point of view and a single one took up less space than multiple small toroids if space is a concern.

I still haven't actually finished mine yet.

Nice! I have another ES9016 board that I have not built yet so I may do something different from stock with that one. I am pretty much an electronics noob so I am stumbling my way through some of this.

Are those Nichicon UKZs? Any reason you did not use the solid polymer caps for all the supplies? Really just looking for some insight on capacitor selection.

Probably sticking with the NE5532 as I do not really have the capability to measure any improvements. I tried OPA1612s but got roughly the same measurements as the NE5532s and found I would get crazy elevated distortion depending on the cables I used for measurement. I assumed this is some sort of instability caused by capacitive loading but as I mentioned previously this is a bit of out of my wheelhouse so I figured it would be safer to stick with stock.

Michael

 

[fluid]

Are those Nichicon UKZs? Any reason you did not use the solid polymer caps for all the supplies? Really just looking for some insight on capacitor selection.

Yes they are UKZ "Fine Gold". I can't exactly remember my thought process as it was late 2016 when I bought the parts. Probably price as the solid Polymers can be expensive and I was unsure if they were ideally suited to the other supplies whereas I knew they were good for the digital line.

Probably sticking with the NE5532 as I do not really have the capability to measure any improvements.

The LM4562 (not 4572 as I wrote above) is a very good stable opamp that is not too expensive, you could try those as they are a step up from the 5532's but not as difficult as the OPA parts.

 

[mdsimon2]

@fluid I see an Arduino in your pictures, I assume you used it to implement hifiduino code? If so any chance you have a copy of the code you used? After looking at the hifiduino site for many months I went to download the code yesterday and it no longer seems to be there.

Michael

 

[fluid]

The pages all still seem to be there for me

https://hifiduino.wordpress.com/introduction-and-guide-to-hifiduino/

Code

https://hifiduino.wordpress.com/code/

You will need the b11f version the links are named as pdf files so you need to download them and change them to .ino

I have not actually had the ES9016 talk to the arduino, the project got shelved before I tested it. It should work there will be an identifier for the ES9016 that is different to the 9018 that will need to be changed just as it was for the 9018k2m. I think I did work out what it was but this is all 5 years ago now so I'll have to dig through and look.

 

[mdsimon2]

The pages all still seem to be there for me

https://hifiduino.wordpress.com/introduction-and-guide-to-hifiduino/

Code

https://hifiduino.wordpress.com/code/

You will need the b11f version the links are named as pdf files so you need to download them and change them to .ino

I have not actually had the ES9016 talk to the arduino, the project got shelved before I tested it. It should work there will be an identifier for the ES9016 that is different to the 9018 that will need to be changed just as it was for the 9018k2m. I think I did work out what it was but this is all 5 years ago now so I'll have to dig through and look.

Doh! I feel dumb for overlooking the pdf renaming, was able to download it without issue, just needed to rename it. Thanks for the help.

In the mean time I have been playing around with Arduino control of the miniSHARC as I think that is where I want to do volume control. I made some slight modifications to the code in the thread below so that volume control works with the VOL-FP. The only thing that does not work with the VOL-FP is changing configuration, but it works fine with the remote or in the plugin so I think I can live with that.

https://www.minidsp.com/forum/opendrc-series-support/10826-readout-of-volume?start=60

I also bought some 4562s but when I started to test them I was running in to significant performance differences (+/- 10 dB THD+N) regardless of op amp used simply by reseating connectors between the power supply and the DAC board. As a result I bit the bullet and directly soldered the regulators to the DAC board and directly soldered the regulators to the SilentSwitcher. I ran some preliminary tests and was getting rock solid THD+N performance regardless of channels used, unfortunately my MacBook ran to in to issue in the middle of testing (touchpad and keyboard stopped working) so I was unable to save the results. Should have the MacBook fixed this weekend and will post some results.

Also bought a cheap die cast aluminum enclosure on amazon to house everything, it is just about the perfect size and was < $40.

https://www.amazon.com/gp/product/B005T78M2W/ref=ppx_yo_dt_b_asin_title_o00_s00?ie=UTF8&psc=1

Michael

 

[mdsimon2]

Replaced the MacBook trackpad/keyboard ribbon cable and we are back to measuring!

Measurements below are with Silent Switcher PSU and LT3045 regulators for DVCC and AVCC with all cabling directly soldered and 4 channels driven using the miniSHARC. As you can see there is no longer the high frequency noise increase that was previously seen with the Silent Switcher and performance is some of the best measured yet, even with multiple channels driven. I've tested a few op amps at this point and have come to the conclusion that apart from stability issues I do not have the hardware to appropriately differentiate.

NE5532 - stock option

LM4562 - maybe slightly better than the NE5532 but difficult to say for sure, not shown in this screen capture but it seemed like they had some odd spikes (still very low level) popping up between 200 and 300 Hz, no stability issues

NJM2068DD - bought these as they have reputation for low noise and was hoping they would solve the high frequency noise rise I was seeing with the Silent Switcher, of course it turned out to be a power supply issue , does seem like they may have a slight noise performance advantage, no stability issues

OPA1612 - with the directly soldered cabling these seem a bit more stable but figured I would show what they look like both when they are OK and when they have stability issues

OPA 1612 - longer cable, stability issues

Michael

 

[mdsimon2]

Also got CamillaDSP up and running with the other ES9016 DAC where I am using the DIYINHK XMOS. I used a RPi 4 running Ubuntu Server 20.04 and ALSA. Definitely took a bit of tinkering but I have Airplay (shairport-sync) and squeezelite running with 8 channel channels of DSP without issue. So far I am doing everything with manual configuration files as I wanted to understand the nuts and bolts of how Camilla operated before considering a GUI option such as Moode. CamillaDSP is super flexible and I can see the advantage of software DSP. That being said the DIYINHK XMOS/DAC seems to have issues with the RPi 4 that I did not see when I was testing with the MacBook. In particular there are some loud squealing noises (of differing duration and volume) when the RPi 4 is rebooted (even though DAC/XMOS stay powered through separate power supply and do not power cycle). This is probably OK for a DIY option that follows proper DAC / amplifier sequencing but probably precludes me from developing this in to a more finished option as it is not very wife friendly.

Michael

 

[mdsimon2]

See below for a quick comparison of miniSHARC IIR filters and CamillaDSP IIR filters. As you can see the results look identical which is nice as it means that miniDSP and CamillaDSP are using the same Q definitions and if you have implemented filters in miniDSP you can apply those without modification in CamillaDSP.

I've also noticed that the above mentioned restart squealing noises with the RPi4 seem to go away if I use an USB 2 port instead of an USB 3 port, now I just get a very benign click when restarting the RPi 4.

One thing that you do need to think about when using CamillaDSP is how you resample to the CamillaDSP sample rate. If you let ALSA do the resampling the results are not good (20+ dB THD+N performance degradation). Post #34 in the thread linked below has some measurements showing the difference between ALSA resampling and SOX resampling (in squeezelite).

https://www.audiosciencereview.com/...nterface-motu-m4-phenomal-dsp-streamer.24493/

Overall pretty pleased with this RPi 4 streamer setup running LMS. A 4 GB RPi 4 + power supply adds about $70 to the cost so for $350 you are looking at a DSP + DAC solution with more power than the more expensive miniSHARC based solution and does not require you to solder a new clock. I am considering adding a digital input HAT such as the HifiBerry DIGI + I/O or DAC+DSP as that would give even more input flexibility. As it currently stands I can airplay to the streamer using the shairtunes2 plugin in LMS which gives a ton of flexibility (can stream Amazon Music HD from an iPhone for example), given the DIYINHK DAC performance is not much better than 16 bit (-96 THD+N) you are really not looking at much performance degradation from airplay which is limited to 44.1 kHz / 16 bit. Of course I can also play files using LMS without any performance degradation as well.

Michael

 

[Kaameelis]

A few times on this board and specifically recently in the ASR open source streamer thread I have seen folks express interest in the DIYINHK multichannel DACs based on the ES9016 and ES9038pro chips. From reading about these DACs on DIY Audio I have seen two groups of comments, 1) implementation is not great (especially the early 2 layer boards) and 2) some folks have built the ES9016 and were quite happy with it. Thought I would try to bridge the two by making some measurements and listening. I have some basic test equipment (scope, Motu M4, DMM) and have been gaining some experience with making measurements. More than happy to experiment so if there is something you want to see feel free to ask.

Why you selected ES9016 but not ES9038pro kit form DIYINHK?

 

[mdsimon2]

Why you selected ES9016 but not ES9038pro kit form DIYINHK?

I did, I commented on it in the second post. Measured performance was not as good as the ES9016.

"I did also purchase the 9038pro based DAC but honestly had lower expectations for that DAC as I understand properly implementing a 9038pro is difficult. This DAC has a local oscillator installed and operates asynchronously. Measured performance is worse than the ES9016 and shows different performance depending on channel. Channel 1/2 and 7/8 (edges of DAC board) have a SINAD of 92 dB and Channels 3/4 and 5/6 have a SINAD of 98 dB (middle of DAC board). I also experienced intermittent noise with this DAC that I do not quite understand. With the base XMOS everything seemed fine, however when using the miniSHARC I would often get noticeable noise, especially when using all 8 channels. The weirdest thing is that once this occurred even if I switched back to the XMOS I would still get noise! The only way to solve it was to shut off the DAC for several minutes and start over. It is possible I made a bad solder joint on the board or something else is damaged but I haven't found the culprit yet. To be honest I am not that interested in pursuing the ES9038pro further as the ES9016 has good measured performance and is rock solid. Would definitely be interested if anyone has played around with the 9038pro board and has experience to share."

Michael