Hello Everyone,
This is a complementary review of the SMSL DO400.
Preamble
This exact item has already been reviewed by Amir, and I asked him to send it to me. I wanted to verify my ability to reproduce meaningful tests as I planned to review more of high precision devices in the future.
Reviewing the exact same DAC is a great opportunity to highlight the differences between the big AP 555x measurements and my modest set of Cosmos ADCiso, Scaler and Notch, as a mean to calibrate them.
SMSL DO400 - Presentation
I'll be quick since you already know this DAC. It was of particular interest to me as it hosts the state of the art ESS9039S, and SMSL is known to master the implementation of that type of DAC. The review of Amir confirmed this is a SOTA DAC.
The back shows all connection we need as a DAC, plus the Bluetooth antena.
User Experience
I like the fact the remote control is not mandatory to operate the SMSL DO400, and the interface is intuitive. Other than that, it feels nice and the display is big enough for me, as opposed to many others.
The SMSL offers a preamp and DAC mode, the former allowing adjustment of the rear analog outputs. I recommend using it since it does not reduce the performances and allow for a better resistance to intersample overs, as I'll show in the review.
When in preamp mode, the 0dB volume display means 4Vrms at balanced outputs (2Vrms unbalanced) and it goes up to +2dB, for 5Vrms output (2.5Vrms unbalanced).
SMSLS DO400 - Measurements
All measurements performed with an E1DA Cosmos ADCiso (grade 0) in mono mode, the Cosmos Scaler (100kohms from unbalanced input) and APU (when required) for analog outputs.
From XLR out (non inverting) I measured:
----
Of course the first test is the standard dashboard of THD+N which determines the SINAD. In the APx555 this is done from either a 400Hz or 1kHz sine tone. Amir uses the 1kHz, as you all know.
On the AES side, it is recommended to use a 997Hz sine tone, for THD+N measurements. The AES also recommends to use dither on max 1LSB (TPDF) although I think it is unnecessary with a sine that is a prime number. I already explained and showed why in the review of the SMSL PO100 Pro.
But as of now, I'll stick to 1kHz too, with dither. However, I won't use the span of Amir (10Hz to 22.4kHz) but the more standard 20Hz to 20kHz, simply because it's been my standard for some time and does not change much to the results.
Here we go, 1kHz @0dBFS with volume set to 4Vrms (48kHz SR, FFT 32k, 4 averages):
The dashboard shows the best channel, the other one has a THD+N of -127.6dB, meaning an average SINAD of 127.65dB. When expanding the span as Amir does, from 10Hz to 22.4kHz, I get an average SINAD of 127.35dB, only 0.3db difference. This is an incredible performance.
Increasing the output volume to 5Vrms got me to 128.35dB SINAD (10Hz to 22.4kHz) and again 0.3dB better with a span of 20Hz-20kHz.
These are crazy numbers, and thanks to the Cosmos APU that notches the fundamental by 30dB in analog domain to ease the life of the ADC (actually same mechanism as in the AP).
Without the analog notch, at 4Vrms, I got 124dB SINAD, still a fabulous result. With that, I know I have enough room to measure super high resolution DACs, which is exactly what I wanted to know!
From RCA outputs (2Vrms), I got an average SINAD 122.15dB, and 123.7dB at 2.5Vrms.
----
From experience with reviewing CD Players, I like to run the same 1khz test 6dB lower:
The two channels are now identical at -122.4dB of THD+N, crazy results again.
----
Stereophile was often showing the below and I like it too, so I'm used to reuse it. It is a view from 20Hz to 1kHz with a 1kHz sine at 0dBFS:
This view allows a better viewing at the power supply potential related noise (leakage) and here there's nothing. I had to do 0 effort to get there, meaning the SMSL is well isolated on power supply perspective. Note that the vertical scale goes down to -170dBr, else there's barely anything to see!
----
Since I'm here, let me add the 50Hz @0dBFS into 600ohms, for compatibility with Stereophile measurements:
This is H3 dominated at a very low -117dBr, very good!
----
There is a selection of oversampling filters, and some are really flat in audio band, but not all of them. The below is a wideband analysis of white noise @-10dBFS, volume set at 0dB, with 44.1kHz input sampling rate:
I recommend using the Linear Fast filter, especially with digital source at 44.1kHz sampling rate.
Note the purple trace, that is with the option "No Filter". Not only it creates a huge number of aliases beyond audio band, it goes well above 0dB, and I guess this is how I burnt my scaler, sending twice more voltage than anticipated into it. This is very bad, and let me show you how bad it goes.
This is again 1kHz @0dBFS with the filter option set to "FILTER OFF", and I lowered the volume to -3.5dB
When setting PCM FILTER = FILTER OFF, the output volume is 6dB higher. And it massively clips, probably the output stage and/or the delta-sigma modulator, I don't know. Things get worse if I increase the volume. Not only this pseudo NOS is useless, but it is very badly implemented here, so why?
----
And since I showed all filters and their respective effects, let me talk about the "Sound Color" options. There are 10 of them, one "Standard", and 3 each for "Rich", "Tube" and "Crystal". In the below traces, I overlaid the 3 options each time, since they just increase the THD:
Rich increases both H2 and H3:
Tube focuses on H2 only:
And Crystal on H3:
As such, SMSL is leveraging the tuning capabilities offered by the ESS DAC. Several registers allow for fine (or rude here) tuning of H2 and H3 that SMSL have leveraged to create extreme distorsion profiles.
Of course this has an impact on all types of distortion, the below is an IMD SMPTE test (zoomed around the second test tome of 7kHz):
If the NOS mode is useless and even dangerous, these sound colors options are potentially fun to play with, on educational perspective. Can you hear that?
----
Let's go back to good news with the Jitter Test (48kHz, 512k FFT. 32 averages):
This is a fantastic trace. The low frequency "stressor" component of the jitter test can be (very) well seen at low frequency, meaning very low noise floor. The two minuscule side-bands around the fundamental are a at staggeringly low -157dBr!
This was with the DPLL set to minimum. It is an additional feature of the ESS9039 that is exposed by SMSL in the on-screen menu. At max level, there are little more side bands meaning it sacrifices a bit of jitter performance, but is supposed to ensure a more stable stream with poor source clock.
----
Of course, I won't forget the Multitone test (48kHz SR, 256k FFT, 16 averages):
From roughly 22bits to more than 23bits distortion-free range!
----
Next is the THD+N vs Frequency (192kHz, 512k FFT 4 averages, 3PPO (20Hz to 20kHz), 4Vrms output):
This test is hitting the max resolution of the ADC. The red trace is performed with the fundamental closer to 0dBFS input, but that stresses the ADC at low and high frequencies, increasing the harmonic distortion that takes over the very low noise. When setting the scaler 6db lower to release the stress on the ADC, it inevitably increase the noise floor relative to the fundamental. But as you see from the plot at 1kHz, we get -118dB THD+N at 1khz, with all noise included up to 90kHz. What a performance! The other channel overlays perfectly, so I did not include it.
----
Last and not least, the IMD SMPTE, which can't be compared to the one of Amir because this measurement is based on FFT analysis of IMD distortion only (192kHz, 128k FFT 4 averages, 1dB step, scaler in auto mode):
This test, completely removing the noise, shows a little hump from -40dBFS to -15dBFS. But with an intermodulation distortion as low as -100dBr from already -45dBFS output (0dBFS=4Vrms), this is extremely low intermodulation distortion.
----
Dynamic range is extremely good, be it CCIR-2K weighted or A-Weighted (0dBFS = 4Vrms):
----
And now let's have a look at other less usual measurements. The one that I like is the resistance to intersample-overs.
This test is performed with dedicated test tones with a specific phase shift to generate digital overs. If anything in the digital path (oversampling filter /DSP / ASRC) does not have enough headroom for digital overs, it will generate clipping within the interpolator.
For that, I'm using a test tone of 11'025Hz with a phase shift of 45° that goes up to +3.01dB over, by 1dB steps. This allows me to process what's the headroom of the digital interpolator of the DAC. Here are the results (THD+N measurement up to 90kHz):
The above means that at full output power (5Vrms from XLR), the interpolator of the SMSL (Linear Fast filter) has a headroom of only 1dB. But because the volume control is done in digital domain, before interpolation, there is more than 3dB headroom when lowering the volume by... 3dB.
So, with volume set at -1dB, you get the best results I've measured so far (my previous best was from the Topping DX1 II). If you use this one as a DAC, I recommend reducing the volume that way, since so many master are too hot. The penalty on SINAD is negligible, especially considering how good it is at 4Vrms.
----
Other measurements (not shown):
Crosstalk is ultra low, even at 10khz.
Pitch error is a very small -2ppm, and takes 2 hours to come to that value (the pitch is -4ppm when the DO400 is cold).
SMSL DO400 - Measurements - Headphones
In high gain mode, I got an already very high 9Vrms with volume set at -7dB (into 200kΩ), meaning a max 25Vrms.
In low gain mode, at 0dB volume, I got a high 4.055Vrms output (into 200kΩ) from the 4.4mm jack.
In low gain mode, 4Vrms output, the 1kHz @0dBFS is the below:
At 50mVrms output, in low gain mode, I get a little disappointing SNR of 84.6dB:
It concurs with the finding of Amir, but my measurement is into 200kΩ not 600Ω. I'll update as soon as I receive the necessary connectors/cables.
Conclusion
I am really happy to know I can measure such a high resolution device.
My personal objective was to set a baseline for future reviews. Now that I'm there, I can do more
The overall performances of the SMSL DO400 are really impressive, from XLR and RCA outputs. This DAC can resolve more than 20bits even with volume down by 3dB, which provides a very nice resistance to intersample overs.
The « sound color » options are a funny experience to get to know if you can hear a bit of H2/H3 distortion into music. Just avoid the stupid « no filter » mode, and I hope SMSL can fix the massive distortion that comes from it near full scale.
I hope you enjoyed this review, and I will certainly update it in the future.
Cheers
Flo
PS: about the crazy SINAD results, I had to verify they were not artificial, from the Cosmos APU that I'm now using. For that, I simply reused a previously reviewed Marantz NA-11S1 which had a SINAD of 110dB, clearly not putting under stress my ADC, even without the APU (analog notch). And with it, I got the exact same result, meaning the APU is well calibrated when I introduce it in the loop, and only allows for the last dB(s) of THD+N/SINAD chase.
This is a complementary review of the SMSL DO400.
Preamble
This exact item has already been reviewed by Amir, and I asked him to send it to me. I wanted to verify my ability to reproduce meaningful tests as I planned to review more of high precision devices in the future.
Reviewing the exact same DAC is a great opportunity to highlight the differences between the big AP 555x measurements and my modest set of Cosmos ADCiso, Scaler and Notch, as a mean to calibrate them.
SMSL DO400 - Presentation
I'll be quick since you already know this DAC. It was of particular interest to me as it hosts the state of the art ESS9039S, and SMSL is known to master the implementation of that type of DAC. The review of Amir confirmed this is a SOTA DAC.
The back shows all connection we need as a DAC, plus the Bluetooth antena.
User Experience
I like the fact the remote control is not mandatory to operate the SMSL DO400, and the interface is intuitive. Other than that, it feels nice and the display is big enough for me, as opposed to many others.
The SMSL offers a preamp and DAC mode, the former allowing adjustment of the rear analog outputs. I recommend using it since it does not reduce the performances and allow for a better resistance to intersample overs, as I'll show in the review.
When in preamp mode, the 0dB volume display means 4Vrms at balanced outputs (2Vrms unbalanced) and it goes up to +2dB, for 5Vrms output (2.5Vrms unbalanced).
SMSLS DO400 - Measurements
All measurements performed with an E1DA Cosmos ADCiso (grade 0) in mono mode, the Cosmos Scaler (100kohms from unbalanced input) and APU (when required) for analog outputs.
From XLR out (non inverting) I measured:
- At 0dB Preamp setting: 4.162Vrms from right channel and 4.153Vrms from Left one, i.e. 0.02dB channel imbalance.
- At +2dB Preamp setting: 5.241Vrms from right channel and 5.230Vrms from Left one, i.e. 0.02dB channel imbalance.
- At 0dB Preamp setting: 2.018Vrms from right channel and 2.013Vrms from Left one, i.e. 0.02dB channel imbalance.
- At +2dB Preamp setting: 2.540Vrms from right channel and 2.534Vrms from Left one, i.e. 0.02dB channel imbalance.
----
Of course the first test is the standard dashboard of THD+N which determines the SINAD. In the APx555 this is done from either a 400Hz or 1kHz sine tone. Amir uses the 1kHz, as you all know.
On the AES side, it is recommended to use a 997Hz sine tone, for THD+N measurements. The AES also recommends to use dither on max 1LSB (TPDF) although I think it is unnecessary with a sine that is a prime number. I already explained and showed why in the review of the SMSL PO100 Pro.
But as of now, I'll stick to 1kHz too, with dither. However, I won't use the span of Amir (10Hz to 22.4kHz) but the more standard 20Hz to 20kHz, simply because it's been my standard for some time and does not change much to the results.
Here we go, 1kHz @0dBFS with volume set to 4Vrms (48kHz SR, FFT 32k, 4 averages):
The dashboard shows the best channel, the other one has a THD+N of -127.6dB, meaning an average SINAD of 127.65dB. When expanding the span as Amir does, from 10Hz to 22.4kHz, I get an average SINAD of 127.35dB, only 0.3db difference. This is an incredible performance.
Increasing the output volume to 5Vrms got me to 128.35dB SINAD (10Hz to 22.4kHz) and again 0.3dB better with a span of 20Hz-20kHz.
These are crazy numbers, and thanks to the Cosmos APU that notches the fundamental by 30dB in analog domain to ease the life of the ADC (actually same mechanism as in the AP).
Without the analog notch, at 4Vrms, I got 124dB SINAD, still a fabulous result. With that, I know I have enough room to measure super high resolution DACs, which is exactly what I wanted to know!
From RCA outputs (2Vrms), I got an average SINAD 122.15dB, and 123.7dB at 2.5Vrms.
----
From experience with reviewing CD Players, I like to run the same 1khz test 6dB lower:
The two channels are now identical at -122.4dB of THD+N, crazy results again.
----
Stereophile was often showing the below and I like it too, so I'm used to reuse it. It is a view from 20Hz to 1kHz with a 1kHz sine at 0dBFS:
This view allows a better viewing at the power supply potential related noise (leakage) and here there's nothing. I had to do 0 effort to get there, meaning the SMSL is well isolated on power supply perspective. Note that the vertical scale goes down to -170dBr, else there's barely anything to see!
----
Since I'm here, let me add the 50Hz @0dBFS into 600ohms, for compatibility with Stereophile measurements:
This is H3 dominated at a very low -117dBr, very good!
----
There is a selection of oversampling filters, and some are really flat in audio band, but not all of them. The below is a wideband analysis of white noise @-10dBFS, volume set at 0dB, with 44.1kHz input sampling rate:
I recommend using the Linear Fast filter, especially with digital source at 44.1kHz sampling rate.
Note the purple trace, that is with the option "No Filter". Not only it creates a huge number of aliases beyond audio band, it goes well above 0dB, and I guess this is how I burnt my scaler, sending twice more voltage than anticipated into it. This is very bad, and let me show you how bad it goes.
This is again 1kHz @0dBFS with the filter option set to "FILTER OFF", and I lowered the volume to -3.5dB
When setting PCM FILTER = FILTER OFF, the output volume is 6dB higher. And it massively clips, probably the output stage and/or the delta-sigma modulator, I don't know. Things get worse if I increase the volume. Not only this pseudo NOS is useless, but it is very badly implemented here, so why?
----
And since I showed all filters and their respective effects, let me talk about the "Sound Color" options. There are 10 of them, one "Standard", and 3 each for "Rich", "Tube" and "Crystal". In the below traces, I overlaid the 3 options each time, since they just increase the THD:
Rich increases both H2 and H3:
Tube focuses on H2 only:
And Crystal on H3:
As such, SMSL is leveraging the tuning capabilities offered by the ESS DAC. Several registers allow for fine (or rude here) tuning of H2 and H3 that SMSL have leveraged to create extreme distorsion profiles.
Of course this has an impact on all types of distortion, the below is an IMD SMPTE test (zoomed around the second test tome of 7kHz):
If the NOS mode is useless and even dangerous, these sound colors options are potentially fun to play with, on educational perspective. Can you hear that?
----
Let's go back to good news with the Jitter Test (48kHz, 512k FFT. 32 averages):
This is a fantastic trace. The low frequency "stressor" component of the jitter test can be (very) well seen at low frequency, meaning very low noise floor. The two minuscule side-bands around the fundamental are a at staggeringly low -157dBr!
This was with the DPLL set to minimum. It is an additional feature of the ESS9039 that is exposed by SMSL in the on-screen menu. At max level, there are little more side bands meaning it sacrifices a bit of jitter performance, but is supposed to ensure a more stable stream with poor source clock.
----
Of course, I won't forget the Multitone test (48kHz SR, 256k FFT, 16 averages):
From roughly 22bits to more than 23bits distortion-free range!
----
Next is the THD+N vs Frequency (192kHz, 512k FFT 4 averages, 3PPO (20Hz to 20kHz), 4Vrms output):
This test is hitting the max resolution of the ADC. The red trace is performed with the fundamental closer to 0dBFS input, but that stresses the ADC at low and high frequencies, increasing the harmonic distortion that takes over the very low noise. When setting the scaler 6db lower to release the stress on the ADC, it inevitably increase the noise floor relative to the fundamental. But as you see from the plot at 1kHz, we get -118dB THD+N at 1khz, with all noise included up to 90kHz. What a performance! The other channel overlays perfectly, so I did not include it.
----
Last and not least, the IMD SMPTE, which can't be compared to the one of Amir because this measurement is based on FFT analysis of IMD distortion only (192kHz, 128k FFT 4 averages, 1dB step, scaler in auto mode):
This test, completely removing the noise, shows a little hump from -40dBFS to -15dBFS. But with an intermodulation distortion as low as -100dBr from already -45dBFS output (0dBFS=4Vrms), this is extremely low intermodulation distortion.
----
Dynamic range is extremely good, be it CCIR-2K weighted or A-Weighted (0dBFS = 4Vrms):
----
And now let's have a look at other less usual measurements. The one that I like is the resistance to intersample-overs.
This test is performed with dedicated test tones with a specific phase shift to generate digital overs. If anything in the digital path (oversampling filter /DSP / ASRC) does not have enough headroom for digital overs, it will generate clipping within the interpolator.
For that, I'm using a test tone of 11'025Hz with a phase shift of 45° that goes up to +3.01dB over, by 1dB steps. This allows me to process what's the headroom of the digital interpolator of the DAC. Here are the results (THD+N measurement up to 90kHz):
| Intersample-overs tests THD+N 20Hz - 90kHz | Peak = -0.99dBFS | Peak = +0.01dBFS | Peak = +1.01dBFS | Peak = +2.01dBFS | Peak = +3.01dBFS |
| SMSL DO400 (Vol +2dB) | -115dB | -115dB | -40.9dB | -25.3dB | -20.4dB |
| SMSL DO400 (Vol 0dB) | -115dB | -115dB | -115dB | -115dB | -40.9dB |
| SMSL DO400 (vol -1dB) | -115dB | -115dB | -115dB | -115dB | -115dB |
The above means that at full output power (5Vrms from XLR), the interpolator of the SMSL (Linear Fast filter) has a headroom of only 1dB. But because the volume control is done in digital domain, before interpolation, there is more than 3dB headroom when lowering the volume by... 3dB.
So, with volume set at -1dB, you get the best results I've measured so far (my previous best was from the Topping DX1 II). If you use this one as a DAC, I recommend reducing the volume that way, since so many master are too hot. The penalty on SINAD is negligible, especially considering how good it is at 4Vrms.
----
Other measurements (not shown):
- First, I now regroup all IMD tests as below, sourced from REW's results (IMD distortion and Total Distortion+Noise shown for each):
- Crosstalk: <-160dB (100Hz), -147dB (1kHz), -130dB (10kHz)
- Pitch Error (GPSDO corrected) : 19'996.96Hz (19'997Hz requested) ie -2ppm
Crosstalk is ultra low, even at 10khz.
Pitch error is a very small -2ppm, and takes 2 hours to come to that value (the pitch is -4ppm when the DO400 is cold).
SMSL DO400 - Measurements - Headphones
In high gain mode, I got an already very high 9Vrms with volume set at -7dB (into 200kΩ), meaning a max 25Vrms.
In low gain mode, at 0dB volume, I got a high 4.055Vrms output (into 200kΩ) from the 4.4mm jack.
In low gain mode, 4Vrms output, the 1kHz @0dBFS is the below:
At 50mVrms output, in low gain mode, I get a little disappointing SNR of 84.6dB:
It concurs with the finding of Amir, but my measurement is into 200kΩ not 600Ω. I'll update as soon as I receive the necessary connectors/cables.
Conclusion
I am really happy to know I can measure such a high resolution device.
My personal objective was to set a baseline for future reviews. Now that I'm there, I can do more
The overall performances of the SMSL DO400 are really impressive, from XLR and RCA outputs. This DAC can resolve more than 20bits even with volume down by 3dB, which provides a very nice resistance to intersample overs.
The « sound color » options are a funny experience to get to know if you can hear a bit of H2/H3 distortion into music. Just avoid the stupid « no filter » mode, and I hope SMSL can fix the massive distortion that comes from it near full scale.
I hope you enjoyed this review, and I will certainly update it in the future.
Cheers
Flo
PS: about the crazy SINAD results, I had to verify they were not artificial, from the Cosmos APU that I'm now using. For that, I simply reused a previously reviewed Marantz NA-11S1 which had a SINAD of 110dB, clearly not putting under stress my ADC, even without the APU (analog notch). And with it, I got the exact same result, meaning the APU is well calibrated when I introduce it in the loop, and only allows for the last dB(s) of THD+N/SINAD chase.
Last edited:
It’s basically the same DAC. So, I’ll pay close attention to the labeling when testing filters.
).
Right, the Hubble telescope isn't AP either.
