tda1387 dac pcb "front end"

Member
Joined 2008
Paid Member
Update on the balanced RPI HAT design: it appears it generates some low-level hiss. With lower-sensitivity amps and/or low-sensitivity speakers, you may not notice. But I recently paired the DAC with a new amp. Compared to what I was using, this amp appears to be a bit more sensitive to source quality. The result was that during quiet passages of less-than-perfect recordings, the hiss was sometimes noticeable enough to be distracting. See this post for more info.

So looks like my next project is to try to clean up some base noise on this DAC. (Hmmm, what about pre-I/V filtering? I wonder if anyone's tried that? ;))
 
Finally had some time to do a quick setup of the SE HAT today.


Rpi3b + kali - piCorePlayer squeezelight and LMS - through mediocre workshop amp and speakers.



Very listenable and no grounding problems etc.



Next a more permanent case and upgraded power supply, as have JRiver will set that up on tinkerboard and move to our everyday system, for some proper listening comparisons.
 

Attachments

  • DSC_1701.jpg
    DSC_1701.jpg
    422 KB · Views: 395
I have been toying around with the Garman 1387 hat for the last several weeks. Greg Stewart kindly loaned me a treasure trove of 5V PSU goodies to try in my Pi stack. He encouraged me to try powering each board separately, and I must admit, each time I added a PSU, clarity went up another notch. Here's my current stack using 4 PSUs. (Temporarily the DAC is jumpered to the Kali to avoid sharing power between them--until I remove that 0 ohm resistor on the Kali to defeat this feature)

Top to bottom:
Gstew modded 5V Jameco linear wall wart -> Garman 1387 hat
Gstew modded 5V Jameco linear wall wart -> Allo Kali
Gstew modded 5V Jameco linear wall wart -> IanCanada IsolatorPi
IFi iPower -> Raspberry Pi
All PSUs are going through a shared CLC via Greg's power conditioner plug.

So far I prefer the sound of Greg's 5V wall warts to everything else I have tried on each level of the stack. This includes: an additional iPower, a 12V linear wall wart into an eBay LT3042 LDO, and some of Greg's modded 12V warts into Acko AK375 LDOs with supercaps added. (Today Greg suggested that I may not have warmed these supplies up enough go to realize full performance.) And also the iPower into a Jan Didden Silent Switcher--probably my second favorite 5V supply so far. Grounding for any of these supplies is essential to reducing noise/hum. I have wires from the ground rails on both top and bottom of the isolator board going direct to earth.

Anyhow, I am very pleased with the beautiful transparency of the rat's nest you see here. I feel I may be getting somewhat close to the upper limits of Pi-Fi. My next step will be trying to replace the iPower on the Pi with a linear 5V supply capable of 2A output. Even with the isolator in place, the Pi's PSU plays a very audible role IME. I am also doing some comparisons between this and other DACs, including DIAL, LingDAC, ApplePi and hopefully Boss.
 

Attachments

  • IMG_3232.jpg
    IMG_3232.jpg
    1,020.7 KB · Views: 391
  • IMG_3233.jpg
    IMG_3233.jpg
    1,000.5 KB · Views: 365
Member
Joined 2008
Paid Member
Finally had some time to do a quick setup of the SE HAT today....

Awesome! Glad you got it up and running. Always encouraging to see my "babies" all grown up and starting their own lives. ;)


My next step will be trying to replace the iPower on the Pi with a linear 5V supply capable of 2A output. Even with the isolator in place, the Pi's PSU plays a very audible role IME. I am also doing some comparisons between this and other DACs, including DIAL, LingDAC, ApplePi and hopefully Boss.

Although the RPi can technically draw up to 2A, if you are using it purely to shuffle PCM bits onto the I2S bus, power utilization should be well under 2A. I don't know what power-saving tweaks Volumio does by default, but in general, with Linux on the RPI, it's not terribly hard to disable hardware you're not using, e.g. HDMI, onboard sound chip, wifi (if using RPI3), etc. The CPU frequency can also be changed, and I have mine fixed to the lowest possible speed.

The Raspberry Pi Zero is somewhat appealing in this regard, except that it lacks a builtin Ethernet controller.
 
Hi SE,


Could you elaborate on the mods done by Greg to those Jameco adapters, any particular reason why are they superior to the others you have auditioned.


I see your LingDac in the pictures, time permitting I too am muddling through a 'Pi-Fi' iteration, will be interesting to read your comparison once you have viewed all your Pi-Fi options.


Best of luck and looking forward to your observations.


Matt - haven't had much time with your 'baby' hence my reserved comments. At the very least I now have a small taste of the potential of the 1387 you all talk about, indeed this is going to be very enjoyable -


Thanks again all pioneers I too hope to become a 'pundit' !!
 
When I experimented with the 1387, I found there is a "correct" voltage for the output node. Where the noise will be minimised.

This happens because of the continous calibration. With each wordclock transition it will calibrate one of various current sources. This current source will be disconnected from the output node (the "spare current source" will be used in its place), calibrated, and reconnected. If the voltage at the output is the same of the internal calibration circuits then there will be no voltage jumps during this process. Thus noise and glitches are minimised. You can figure this from the various philips papers.

Actually, to make this simpler, think of it as finding the correct voltage between output and V+. You can vary the supply voltage to the dac chips instead of changing I/V resistors. Find the point for lowest noise. Important, must be done with all input signals present, of course. With the dac playing silence.

Thanks,
Alex




Update on the balanced RPI HAT design: it appears it generates some low-level hiss. With lower-sensitivity amps and/or low-sensitivity speakers, you may not notice. But I recently paired the DAC with a new amp. Compared to what I was using, this amp appears to be a bit more sensitive to source quality. The result was that during quiet passages of less-than-perfect recordings, the hiss was sometimes noticeable enough to be distracting. See this post for more info.

So looks like my next project is to try to clean up some base noise on this DAC. (Hmmm, what about pre-I/V filtering? I wonder if anyone's tried that? ;))
 
Member
Joined 2008
Paid Member
Hi Alex, thanks for the input!

When I experimented with the 1387, I found there is a "correct" voltage for the output node. Where the noise will be minimised.

Just out of curiosity, what kind of I/V were you using when you experimented with the tda1387?


This happens because of the continous calibration. With each wordclock transition it will calibrate one of various current sources. This current source will be disconnected from the output node (the "spare current source" will be used in its place), calibrated, and reconnected. If the voltage at the output is the same of the internal calibration circuits then there will be no voltage jumps during this process. Thus noise and glitches are minimised. You can figure this from the various philips papers.

Makes sense. I'm curious, then, if you did any experiments with changing the sample rate? If the calibration happens on every wordclock, then changing the sample rate of the source material out to result in a corresponding change in re-calibration rate, which should in turn affect the noise. Presumably measurable, I'm curious if it's audible? I might give this a try just because it's easy.


Actually, to make this simpler, think of it as finding the correct voltage between output and V+. You can vary the supply voltage to the dac chips instead of changing I/V resistors. Find the point for lowest noise. Important, must be done with all input signals present, of course. With the dac playing silence.

In my current RPI HAT designs, it's hard to change supply voltage because I'm getting it from the RPI (5v). I'm also using this very simple common-base active I/V from Abraxalito. See attached schematic image.

BREF is the reference voltage for the transistor base, and is generated from a TL431 used as a shunt regulator. With a BC807, BREF=2.8v. I've also used the SSM3J327R MOSFET here, with BREF=2.5v.

Do you have any suggestions on what could be tweaked here to possibly reduce/remove the noise?

Thanks!
 

Attachments

  • tda1387_common_base_iv-20180524.jpg
    tda1387_common_base_iv-20180524.jpg
    69.8 KB · Views: 364
Matt,

I was using a simple passive I/V. I found that a carbon resistor gave me a more pleasant sound. Removed some edginess and "metallic" characteristics. It was around 2K.

I had a strong preference for higher sample rates. The residual noise was indeed afected, but that was not the only reason I prefered it. It sounded better.

About that I/V, I´m not too fond of it, the base current is one concern. Maybe it isn´t a problem if the hfe is very linear. Will it be linear at such a low current (say 100uA to 1mA)? I´m not sure.
 
Matt, I should say I haven´t tried the simple PNP circuit, I should have tried it when I was experimenting with a single 1387. Maybe I will.

In any case, using a bit of digital volume removes the concern of current going to zero.

So did you perceive any difference in noise with higher sample rates?

An alternative is to tweak the "BREF" voltage on your schematic, it may help you get rid of noise.

-Alex
 
Although the RPi can technically draw up to 2A, if you are using it purely to shuffle PCM bits onto the I2S bus, power utilization should be well under 2A..

Well, I tried it. And...I have had great success powering my Pi with a 1A PSU after all! Boots every time and runs indefinitely. :) I am using another one of Greg Stewart's modified Jameco linear wall warts (4 total on my Pi stack now).

Let me just say that this PSU stomps all over my iFi iPower.

I could STILL hear noise from the iPower on the Pi, even with an isolator board in use! Now, with my entire stack powered by these bargain linear supplies, the noise is gone. The treble is cleaner and more relaxed as well...who knows why. Hard to believe. And it costs $20! Plus the cost of parts to mod it...not sure how much total. But I know Greg is planning to share his Jameco mods soon with a photo tutorial. Good stuff.
 
Member
Joined 2008
Paid Member
Sorry for the radio silence, all. I've taken a break from DACs to work on several side projects. The last one is still on-going, but I'm excited to get back to DAC talk!


Are you using digital volume?

Yes, I am.


Matt, I should say I haven´t tried the simple PNP circuit, I should have tried it when I was experimenting with a single 1387. Maybe I will.

So did you perceive any difference in noise with higher sample rates?

An alternative is to tweak the "BREF" voltage on your schematic, it may help you get rid of noise.

I haven't had a chance to try higher sample rates. (Keep forgetting!)

Abraxalito helped me arrive at the BREF values. I can explain from rote how the BREF value was derived, but my understanding of this is superficial at best. First, I'm powering the tda1387 chip(s) with 5v. Per the DS, that results in output voltage compliance of 3.5v. Best to quote Abraxalito on the rest:

Abraxalito said:
We need to arrange the transistor to give no more than 3.5V on its source (or emitter if bipolar) which means biassing the gate (base) lower by the threshold voltage (or a diode drop for bipolar). With MOSFETs the gate threshold voltage is quite variable, much more variable than for a base-emitter junction. The J327 has a very low threshold voltage for a MOSFET which is one thing making it suitable for this application, its maximum is 1V (at 1mA drain current) but the min is 0.3V. So we need to set the gate voltage to accommodate the worst case (1V) threshold and this means setting the gate to 2.5V

That was in the context of using a SSM3J327R MOSFET for that common-base transistor.

A similar thought process leads to a BREF of 2.8V for a BC807 PNP transistor used in this position.

I have tried both the J327 with BREF=2.5v and BC807 with BREF=2.8v, but I have not yet tried other values for BREF.


I am using another one of Greg Stewart's modified Jameco linear wall warts (4 total on my Pi stack now).

Let me just say that this PSU stomps all over my iFi iPower.

And it costs $20! Plus the cost of parts to mod it...not sure how much total. But I know Greg is planning to share his Jameco mods soon with a photo tutorial.

Did he ever post the details of that?

While you're experimenting with power supplies, have you tried these ultracapacitors as discussed in this thread? They are basically using them like batteries because their capacity is so high relative to the current draw of the application. But so far only positive comments! (Personally, I'm a little scared of them!)
 
I did experiments using ultracaps to power my TDA1387 DACs. Sure they worked great but a fairly big work-up to use them, not to mention the sheer bulk makes them impractical for portable applications. In the end I ditched them when I went over to active I/V as the bass was at least as good with that as for passive I/V plus supercaps.
 
Did he ever post the details of that?

While you're experimenting with power supplies, have you tried these ultracapacitors as discussed in this thread? They are basically using them like batteries because their capacity is so high relative to the current draw of the application. But so far only positive comments! (Personally, I'm a little scared of them!)

No, Greg hasn't shared the Jameco mods yet. I asked for an update but haven't heard from him for a few weeks. I know he is busy playing around with all kinds of digital source gear, including the supercaps (yes I am following Ian's thread too).

After hearing such drastic differences in PSUs in my tests, and reading so many favorable impressions of powering DACs with LifePo4 batteries, I am VERY interested in the supercaps. I just need to learn what circuit is necessary to use them.
 
Hi matt_garman,
Thank you for all of your great work !

I am still enjoying the modified TDA137 that you and abraxalito gave me guidance.
Cayin N3 /Volumio > Raspberry Pi > Hifiberry Digi + > TDA1387 > JLH 1969 Amp > Pioneer SP-BS22-LR Speakers

Now setting up a desktop rig for headphones thinking:
Cayin N3 / Volumio > Raspberry Pi > ALLO - Kali i2s Reclocker > Hifiberry Digi + > TDA1387 > JLH 1969 Headphone Amp > HD650

Do you think your "TDA1387 DAC "front end" would work well in this configuration ?

Thank you !
- Mark
 
Member
Joined 2008
Paid Member
I did experiments using ultracaps to power my TDA1387 DACs. Sure they worked great but a fairly big work-up to use them, not to mention the sheer bulk makes them impractical for portable applications. In the end I ditched them when I went over to active I/V as the bass was at least as good with that as for passive I/V plus supercaps.

I recently re-read some old tda1387 threads, and saw your "hexacap" made with ultracaps. In previous readings, I missed the detail that you used ultracaps (I thought you just paralleled a bunch of regular electrolytics). Do you remember which ultracaps you used? In the other thread, they are using single (or double just to get enough voltage) ultra caps - what I would call ultra-ultra caps because their current/capacity rating is just crazy, so no need (in theory) to parallel them.

At this point, I am curious about ultracaps, but content to watch others experiment with them.


any improvement betwen dual mono and single dac chip?

When you say "dual mono", I assume you mean balanced output. From my perspective, there are a few benefits:
  • It forces the use of an i2s buffer of sorts. In my designs, I'm using the HC86, so I can invert the DATA signal for the "cold" part of balanced output; the word clock and frame size signals are buffered. (You could of course use a buffer for a single DAC chip as well, but you don't have to.)
  • The benefits of differential signaling are well-established: elimination of ground loops, common-mode noise reduction, etc.
  • Some downstream components may prefer (or demand) differential input. For example, the newer TI class-D amplifiers (e.g. tpa3251) require differential input for stereo BTL mode. Most implementations put an op-amp in front of the amplifier input to create the differential signal from a single-ended source. But if you already have a balanced signal, you can by-pass the extra circuitry.
  • Balanced is a superset of single-ended. So even if the DAC has balanced outputs, it implicitly also has single-ended outputs.
  • As Abraxalito has noted, having two DAC chips with their DATA lines interverted better balances the power supply loading.

For all the above reasons, I prefer balanced interconnects. However, if you ask me if I can hear a difference... well, ugh, umm... :eek: ;)


Now setting up a desktop rig for headphones thinking:
Cayin N3 / Volumio > Raspberry Pi > ALLO - Kali i2s Reclocker > Hifiberry Digi + > TDA1387 > JLH 1969 Headphone Amp > HD650

Do you think your "TDA1387 DAC "front end" would work well in this configuration ?

Hi Mark, I haven't really spent any time with the "front end" in quite a while. It would certainly work. But since you are using a Raspberry Pi, I would recommend one of the RPI HAT designs, and optionally put an Allo Kali (I2S reclocker) between the RPI and the DAC. I think this is the easiest/simplest way to get going with the tda1387 (but I'm highly biased!).

That said, I'm thinking about a new design that combines the ideas of the front end with a RPI HAT format. But right now it's just in the "thinking about it" stages.
 
Member
Joined 2008
Paid Member
stellarelephant - on the single-ended v1.3 board (what you have) there are only three total active devices that need power: the tda1387 chip itself, and the two BC560 I/V transistors. The tda1387 DAC chip takes 5.0v, and the BC560 transistors actually expect 2.8V. I use the TL431 to generate 2.8V from 5.0V.

See the attached picture. First thing you'll need to do is cut the copper fill in the upper left (indicated by the orange circle I drew). This is where I get 5V power from the RPI GPIO header. That's a fairly thick copper trace, might take some work to remove it. Then you can probably solder your dedicated 5V PSU leads to the capacitor (C5) leads on the bottom.

Now if you want to use a second dedicated supply for the I/V transistors, you have two options. First (probably easier), is to cut the bottom trace indicated by the orange circle in the middle. That's the trace that provides power to the TL431 (which in turn sets the 2.8V for the I/V transistors). You could then solder your second dedicated 5V PSU leads to that surface mount resistor (R1). (Note the TL431 actually takes a fairly wide "input" voltage, if you happen to have a good PSU that's not necessarily 5V; see the datasheet.)

Your second option, if you have a dedicated 2.8V power supply: still cut that bottom 5V trace, or simply remove R1. But also remove the TL431 (D1). Then solder your 2.8V PSU leads to the bottom of that capacitor (C6). 2.8V is a rather unusual voltage, but there are lots of LDOs that can generate a custom voltage.

As a side note: you might be interested in my next iteration on simple tda1387 DACs: this attempts to combine the idea of a "front end" with an RPI HAT. It's 4x tda1387 chips in a 2x2 balanced arrangement. Has a jumper for RPI power or dedicated 5V power. Has a pin header for I2S input so it can be used as a standalone DAC. Basic passive I/V onboard, or simply omit I/V components and have direct current-out (for a dedicated I/V board). These have been fabbed, just waiting for them to be shipped to me.
 

Attachments

  • tda1387_rpi_hat_dac_v1.3_voltage_mods.jpg
    tda1387_rpi_hat_dac_v1.3_voltage_mods.jpg
    535.5 KB · Views: 309
  • tda1387_rpi_bal_fe_v1.0_3drender_20180811.jpg
    tda1387_rpi_bal_fe_v1.0_3drender_20180811.jpg
    184.8 KB · Views: 283
Thanks for the detailed reply, Matt. The pics are very useful!

Thankfully I won't need to cut the fat GPIO power trace at top left since I already have made provision to send dedicated power to the board by NOT stacking the board atop my Pi stack and instead using GPIO jumpers for I2S and with power jumpers going to separate PSU.

I WILL try my hand at cutting the middle trace to get a dedicated 5V feed for the TL431 and I/V transistors (option #1). I will be using another one of Greg Stewart's modded Jamecos. Greg actually has the DAC for audition at present, so it'll be a week or so before I am able to try this.

I have an LT3042 module that could be used to drop a 5V Jameco to 2.8V, for option #2 down the road. It isn't the best sounding regulator module in the world, but if it is superior to the TL431, perhaps it would give another SQ bump.

Your new balanced Pi Hat looks cool, and the added power and I/V flexibility make it quite attractive. Impressive 3D model too! I have not heard passive I/V yet. Are you intending to use it like this yourself or to follow with a separate I/V board? At any rate I appreciate the modular approach. I have been itching to build the new Pass B1 NuTube buffer...I wonder how that would pair up with passive I/V. All my gear is single ended. What will signal output voltage be for SE and balanced?