don't trash your TEAC UD-H01 DAC, pimp it!

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Now to some quirks I stumbled across, regarding on how Linux / Ubuntu seems to interact with the TEAC, fed from PulseAudio versus ALSA sound layer.

What can be seen on that measurements is, that digital streams from PulseAudio are interpreted by the TEAC one by one.
On signals with high crest factors, such as a square wave for example, hard clipping occurs.

On the other hand TEAC or more probably even ALSA at Linux / Ubuntu incorporates kind a soft clipping mechanism not to allow digital 0dB signals with high crest factor to go into clipping at all.
At around digital -6dB, signals from ALSA still are compressed as can be seen.
Not really bit perfect I'd say.
 

Attachments

  • sine_no_clipping_1kHz_16_44_0dB_PULSEAUDIO.PNG
    sine_no_clipping_1kHz_16_44_0dB_PULSEAUDIO.PNG
    16 KB · Views: 350
  • square_hard_clipping_1kHz_16_44_0dB_PULSEAUDIO.PNG
    square_hard_clipping_1kHz_16_44_0dB_PULSEAUDIO.PNG
    13.1 KB · Views: 343
  • square_hard_clipping_1kHz_16_44_0dB_PULSEAUDIO_zoom.PNG
    square_hard_clipping_1kHz_16_44_0dB_PULSEAUDIO_zoom.PNG
    12.8 KB · Views: 343
  • sine_no_clipping_1kHz_16_44_-6dB_PULSEAUDIO.PNG
    sine_no_clipping_1kHz_16_44_-6dB_PULSEAUDIO.PNG
    13.6 KB · Views: 338
  • square_no_clipping_1kHz_16_44_-6dB_PULSEAUDIO.PNG
    square_no_clipping_1kHz_16_44_-6dB_PULSEAUDIO.PNG
    14 KB · Views: 336
  • square_no_clip_1kHz_16_44_-6dB_ALSA.PNG
    square_no_clip_1kHz_16_44_-6dB_ALSA.PNG
    13.4 KB · Views: 75
  • sine_no_clip_1kHz_16_44_-6dB_ALSA.PNG
    sine_no_clip_1kHz_16_44_-6dB_ALSA.PNG
    12.3 KB · Views: 51
  • sine_soft_clip_1kHz_16_44_0dB_ALSA.PNG
    sine_soft_clip_1kHz_16_44_0dB_ALSA.PNG
    13.4 KB · Views: 69
Last edited:
The last quirk I came across is something that became immediately evident at looking at the scope when playing at different sample rates.
At 44kHz the traces did literally walk and wobble quite obviously.
So I set up for measurements that capture envelopes.
Have fun!

Wobble, or most probably jitter created by synchronous USB streaming, is heavy at 44kHz, better already at 48kHz and best at 192kHz.
For easy comparison the last pix shows 192kHz with envelope mode disabled.

If my interpretation is correct, then what TEAC did advertise for the UD-H01 obviously does not hold.
They stated that incoming digital USB signals always will be received in asynchronous mode, exactly to avoid such jitter.
 

Attachments

  • 1kHz_square_16_44_0dB_ALSA_envelope_mode.PNG
    1kHz_square_16_44_0dB_ALSA_envelope_mode.PNG
    18.3 KB · Views: 73
  • 1kHz_square_16_48_0dB_ALSA_envelope_mode.PNG
    1kHz_square_16_48_0dB_ALSA_envelope_mode.PNG
    15.5 KB · Views: 71
  • 1kHz_square_16_192_0dB_ALSA_envelope_mode.PNG
    1kHz_square_16_192_0dB_ALSA_envelope_mode.PNG
    13.6 KB · Views: 86
  • 1kHz_square_16_192_0dB_ALSA_no_envelope_mode.PNG
    1kHz_square_16_192_0dB_ALSA_no_envelope_mode.PNG
    12.5 KB · Views: 65
Well, there are proven solutions, albeit off topic...

Upsampling isn't really a problem. Can do a batch convert with dBPoweramp to any sampling format I like to force feed the TEAC with very easily.
As for the assumed USB jitter, I am currently digging some deeper by ARTA measurements, but at such sub Hz frequencies its really hard to get meaningful results even if jitter is as pronounced as it seems to be from my scope's measurements.


Got the OPA1620 EVAL board.
What a grotesque waste of resources, shipping such a tiny board over the pond by such big parcels!
Most possibly will try it with balanced signal pic up directly at the motherboard and a DACT style shunt pot for volume control.



Despite the measured quirks mentioned before, the UD-H01 already has very good SQ to offer.
Even with dreadful MP3 web streaming it sounds quite astonishing.
Where the TEAC UD-H01 begins to shine though, is with playback of true highres audio recordings, say 24bit / 192kHz.
Clearly discernible from a 16bit / 44.1kHz red book pressing.
The higher sample rate helps in smoothing out the upper registers and the higher bit rate helps in letting you dive deeper into the presentation, a big step forward towards an analogue feeling in short.

To get there the easiest way, I highly recommend to give mature Daphile a try on a spare notebook or barebone. Daphile was written by a Finn based on ArchLinux.
Daphile – Digital Music Convenience for Audiophiles
https://www.daphile.com/download/DaphileInstallation.pdf
Daphile - Audiophile Music Server & Player OS

You can get it up and working nicely in about half an hour, having at hand the right installation ISO, a second PC in the same network, IP and sub net mask for the router and DNS server.
I got away with an old Lenovo N500 Notebook plus SSD and 1GB ram right "out of the box".
Its best to do installation with the DAC already connected to the USB port, to allow Daphile to recognise it at setup routine. If possible, disable internal soundcards in BIOS before setup.

As installation of Daphile is that easy and straight forward, I even tried on an ATOM CPU based Eee-PC, but possibly would need some deeper research into the advanced audio (buffering) settings to iron out glitches that did occure now and then with such a low horse power setup.

To avoid clipping of hot recordings under any circumstances, I recommend to set "attenuation" in the advanced audio settings to -3dB.

Daphile can be configured to be operated remotely by a second PC, or some apps, or directly at the PC you installed it to, or even by the comfortable Slimdevice Duett controller.
https://www.logitech.com/images/pdf/userguides/eng/Logitech_Squeezebox_Duet-ENG.pdf
 

Attachments

  • Daphile.JPG
    Daphile.JPG
    180 KB · Views: 150
  • TI_resource_waste.JPG
    TI_resource_waste.JPG
    261.8 KB · Views: 123
Last edited:
Upsampling isn't really a problem. Can do a batch convert with dBPoweramp to any sampling format I like to force feed the TEAC with very easily.

With Daphile it "should" not even be necessary to batch convert my library as it can do up-sampling on the fly by itself, either to the max sample-rate or to max sync sample-rate.

Drawback is that it does not work out as one would expect.

In the plot below we see that we do not gain from TEACs improved reconstruction filtering at higher sampling frequencies nor do we gain from the less jitter seemingly associated to that.

The measurement was taken by feeding the TEAC with a by Daphile up-sampled -6dB square from 44.1kHz to 176,4kHz.
TEAC USB DAC does correctly indicate an incoming signal of 176,4kHz.
The scope this time was set for single triggering to also show the weirdness going on with the assumed USB jitter.
 

Attachments

  • weird_jitter.jpg
    weird_jitter.jpg
    120.4 KB · Views: 134
Last edited:
Above measurement was taken with Daphile up-sampling set to use minimum phase response.

Things sort of improve with Daphile up-sampling set to use linear phase response.
With this setting, we at least gain from the better reconstruction filtering of the TEAC DAC at higher sample rates, but its superimposed by pre-ringing, most probably due to Daphile's up-sampling set to use linear phase.

As for the assumed USB jitter one can see from the cursor readout that single elongated periods translate to 980-990Hz instead of spot on 1kHz.
Exactly the same behaviour as with the measurement taken above.
 

Attachments

  • preringing_elonged_periods.jpg
    preringing_elonged_periods.jpg
    94.3 KB · Views: 83
Last edited:
Finally got some decent sound out of the OPA1620 EVAL board driving my AKG K701.

By far the best results I got when connected directly and symmetrically to the I/V MUSES 9820 out. Worse was to feed it by the XLR out and worst was to feed it asymmetrically from that point of the mobo where the original headamp pics up its signal.
In addition to that, one interesting observation was that quite in contrary to the stellar PSRR specs this OPA1620 is very picky for its supply.
A fist attempt with polarized 100uF Nichicon Muse 100uF / 50V as blocking capacitors right at the board showed severely compromised PRAT by an overly liquid presentation.
Switching to 1000uF Nichicon FC parallel by 100nF MKT and some pF silver MICA did the trick.
Not quite as refined and balanced as I would like to have it after a few hours of break in, but now definitely heading into the right direction: high resolution, good attack and way better resolved tails.
Also my subjective feeling is that channel separation isn't all that great as could be expected from specs. Would possibly gain further from a split PS arrangement.

As this reservoir cap trick did work out surprisingly well, I assume the rails simply are inadequate regulated. Done by fix voltage 3-pin regs as far as I can see.

I'll try to do the reservoir cap trick at the I/V MUSES 9820 in a next iteration. Lets see what it brings up.
:)

Another direct consequence of above observations is, that the XLR driver stage has potential to be improved by quite a margin.
 

Attachments

  • IV_signal_pic_up.JPG
    IV_signal_pic_up.JPG
    297.4 KB · Views: 229
  • OPA1620_EVAL_1.JPG
    OPA1620_EVAL_1.JPG
    211.3 KB · Views: 229
  • OPA1620_EVAL_2.JPG
    OPA1620_EVAL_2.JPG
    205.4 KB · Views: 275
Last edited:
Is it that all those 47uF have holes for THT capacitors beneath them? At least it seems so looking at the first pic.

yes, always rotated by 90deg.



Speaking of channel separation, spaciousness and "3D" rendering.

One of the most crazy things I came across recently was about galvanic isolation of the USB DAC data stream from its feeding PC.
No one does have any good 'ol school explanation from an engineering point of view, as to why this could make any difference at all, given the fact that non of the digital data gets lost on transfer and clocking anyway is done at the DAC side.
Nevertheless becoming curious, I decided to simply give it a try, neither going with the scientists camp nor with the esoteric one. Especially as Intona sold some of their ugly, white version plastic boxes at a discount via Ebay.

To much of my surprise, the TEAC UD-H01 easily let me differentiate the sonic footprint of the USB connection, or better the absence thereof, when the Intona USB 2.0 Hi-Speed Isolator (Standard Version 1kV) is up-streams to the DAC. Most apparent with complex, highly dense material as for example from the shown album.
A great way, to crank SQ of that cheap to have DAC another step.

Have fun!
 

Attachments

  • Intona_2.JPG
    Intona_2.JPG
    204.5 KB · Views: 115
  • Intona_1.JPG
    Intona_1.JPG
    221 KB · Views: 154
Last edited:
Oh, you were absolutely correct about the FedEx packing ;)

It's waiting to be tested, not today, its listening music time only...

First testbed: Dual (Mono) PCM1794A board, with ADA4898-2 as I/V and LT1028 for Bal/Unbal and filtering. PCM1794A had some mods (Vcc=8V and Iref=6K8) applied taken from DDDAC 1794 (credits to Doede Douma!) Has some issues with I/V opamp stability...
Balanced out to INA1620EVM after I/V and first filter stage.

Second: Dual (Mono) TDA1541A in parallel (not balanced), currently with ADA4898-2 I/V, some mods taken over from ECDesings (John, big thank you!) Will have to got balanced on that, and waiting for the AD844 (I/V) to arrive. The buffer will be ADA4898-2 heading out to the INA1620EVM

Both fed with either straight I2s (1) or in TDA1541A (2) specific "simultaneous"
mode from a JLSounds USB2I2S VIII interface with isolation from XMOS/USB and reclocking after isolation). Will have to got balanced on that, and waiting for the AD844 (I/V) to arrive. The buffer will be ADA4898-2 heading out to the INA1620EVM


Well, TDA1541A will win worlds over the PCM1794A, that's to be told already ;)

EDIT: I think PCM1794A(3) and PCM1795(4) aren't to big, so even if (4) has only half of the current swing, it might be a thing that you can try to passive I/V and take the NKM Muse as a pure gain stage, then feeding directly the INA1620. Oh yeay, you might have to replace some 0805 resistors and possibly rip of some connections on the PCB :rolleyes:

EDIT2: use K702 an all tests

just my thoughts :)
 

Attachments

  • PC291259.jpg
    PC291259.jpg
    467.8 KB · Views: 457
  • PC291256.JPG
    PC291256.JPG
    450 KB · Views: 396
Last edited:
I was wrong.
Sorry, I was completely wrong in my assumption that the observed and shown jitter behaviour could possibly be related to USB data transfer.
It is not, not at all, as came to me when I compared measurements of the UD-H01 with those I took from a TEAC UD-301.

The UD-301 differs in using shunt regulators (TL431) at specific rails, having a way more elaborated cap free, servo controlled headphone amp, and in the way it gently handles past Nyquist barrier content of artificial signals (squares with lots of high order harmonics) deliberately thrown on it.
What UD-301 and UD-H01 do having in common, is the same DAC chip (TI PCM1795) and more important to the topic at had, that both of them don't feature dual clocks.

Below plots show how the UD-301 handles full scale 1kHz square files at sample rates of 44.1kHz and at 48kHz respectively.
First off, the UD-301 seems to do sort of very effective anti-aliasing processing prior to digital to analogue conversion, resulting in a "better" shaped square output. In normal situations this "should" of course already have been done at the recording process.

Looking at measurements taken in envelope mode brings up, that there is no jitter at 48kHz sampling rate (and multiples thereof) to be observed, quite in contrary to 44.1kHz (and multiples thereof) playback.
This tells me that the jitter in fact does come from a quick and dirty trans-sampling process that simply fills out gaps by inserting single elongated periods every now an then.

What that means is, that we are stuck by digital audio history and its vastly incompatible 44.1kHz versus 48kHz base sample rates and therefore one that isn't at ease with such hefty 20us built in trans-sampling jitter at very low frequencies is in need of a DAC with dual clocks.
Dual clock feature comes with the TEAC UD-503 at earliest as it seems. Though, besides having the nice Asahi Kasei AK4490 DAC chip built in, the UD-503 cuts costs by using a very simplistic "one for all" buffering strategy for all of its outputs, including the headphones out.
 

Attachments

  • teac_ud-301_44Khz_square_0dB.JPG
    teac_ud-301_44Khz_square_0dB.JPG
    97.5 KB · Views: 355
  • teac_ud-301_48Khz_square_0dB.JPG
    teac_ud-301_48Khz_square_0dB.JPG
    95.4 KB · Views: 333
Last edited:
did you some modification on the TEAC UD-301?
No sorry, my former post was not meant for the UD-301. My UD-301 went back untouched.

But as for my listening impression, its stock form is quite a step up to the UD-H01 in stock form IMO mostly due to its more refined design, but not enough to come even close to my modded UD-H01.

So I rather decided not the go the same route again, especially as the UD-301 does not feature a dual clock and any 44.1kHz material (and multiples thereof) suffer from the same trans-sampling jitter as the UD-H01 does.

As for you modding, I would suggest to just do the same as I did with the UD-H01. Roll the capacitors first.
For the UD-301 there is a service manual and schematics to dig up in the web, so its a much easier platform for modding.


Good luck
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.