Switching back to just I/V resistor seemed to fix the noise for me so I assumed it was down to not having a centre tap like you said.
I havent done much since with the TDA1387, getting multichannel I2S output from the PC to allow 1 chip per channel (or 4 chips in total for 4 channel digital crossover) is the next thing in mind. From listening I think channel seperation is what these struggle with most compared to my 9038 DAC, so we can see how much that helps.
I havent done much since with the TDA1387, getting multichannel I2S output from the PC to allow 1 chip per channel (or 4 chips in total for 4 channel digital crossover) is the next thing in mind. From listening I think channel seperation is what these struggle with most compared to my 9038 DAC, so we can see how much that helps.
Last edited:
Presumably at that time filters from NPC or BB's DF1700 - I might be mistaken, but have a faint memory that the BB was (based on the) or was a NPC - later the according enhanced versions.
Philips digitalfilters were already dropped at that time afair.....
So far, the only digital filter compatible with TDA1387 (i.e. I2S output) is Philips' own SAA7220. I had a look at various DSs (BB, NPC, Yamaha) and all that I saw put out data in dual mono mode. That is they have a separate data line for left and right.
For quite some time I've been planning on turning a cheap ($2-$3) ARM SoC into a digital filter, perhaps I'll get around to it this year, who knows? Just recently NXP announced their LPC55XX range using Cortex M33. I got the EVM for the LPC55S69 a few weeks ago - the chip includes a hardware accelerator for doing FIR filtering, not sure yet whether its well suited to audio.
For quite some time I've been planning on turning a cheap ($2-$3) ARM SoC into a digital filter, perhaps I'll get around to it this year, who knows? Just recently NXP announced their LPC55XX range using Cortex M33. I got the EVM for the LPC55S69 a few weeks ago - the chip includes a hardware accelerator for doing FIR filtering, not sure yet whether its well suited to audio.
Yes, additional logic was required for interfacing to the I2S world.
Were the SAA7220 variants still in production when the TDA1387 came out?
With a single TDA operating per channel there was quite a noticeable improvement, Im also using active I/V, the same I/V that I use for my 9038Pro DAC. A standard op amp I/V, only using OPA1622s and driving HPs directly.
The non-inverting input references to ground, like the datasheet, while the sabre DACs use a seperate vref for each channel, which is half the DAC's analogue 3.3V supply. I am not sure about the reason for this specific voltage, someone mentioned that it is to cancel the DC offset from the DAC output, but this does not happen, as me and others have noticed.
I know that one reason for using vrefs instead of common ground is to improve channel seperation, so having gone as far as using 1 TDA per channel this will be the next tweak.
Also, going to back to one chip I think it sounds better than multiple, but on the other hand the stacked chips seemed more technically proficient.
Its cool that you taking different approaches with the designs, because there is probably not a single best option.
The non-inverting input references to ground, like the datasheet, while the sabre DACs use a seperate vref for each channel, which is half the DAC's analogue 3.3V supply. I am not sure about the reason for this specific voltage, someone mentioned that it is to cancel the DC offset from the DAC output, but this does not happen, as me and others have noticed.
I know that one reason for using vrefs instead of common ground is to improve channel seperation, so having gone as far as using 1 TDA per channel this will be the next tweak.
Also, going to back to one chip I think it sounds better than multiple, but on the other hand the stacked chips seemed more technically proficient.
Its cool that you taking different approaches with the designs, because there is probably not a single best option.
Last edited:
I had inconclusive results in paralleling chips before, before using transformers that is. With the trafo there's no doubt in my mind that the bass is improved - there's more low-frequency ambience evident, the soundstage is more detached from the speakers. Also evident is a bit more 'drive' to the music, similar to how a classA amp sounds when compared to a classAB.
When using opamp I/V to me its highly plausible that multiple chips sound worse - the extra output current is making the opamp work harder and that puts more load-induced ripple on the supply. What the trafo brings to the party is the ability to have just the right amount of output current from the DAC stage, to match the I/V stage. If using a voltage feedback opamp (the usual type) then a lower output current I reckon will sound better as there's flexibility with the selection of the feedback R.
When using opamp I/V to me its highly plausible that multiple chips sound worse - the extra output current is making the opamp work harder and that puts more load-induced ripple on the supply. What the trafo brings to the party is the ability to have just the right amount of output current from the DAC stage, to match the I/V stage. If using a voltage feedback opamp (the usual type) then a lower output current I reckon will sound better as there's flexibility with the selection of the feedback R.
Be happy to explain all that stuff if you like over in the ES9038Q2M board thread.
Ah interesting, didnt think about that. So a larger value is preferred for feedback to some extent? I noticed with 9038's fairly high OP current that using higher value resistors tended to sound better. I thought this was the DAC sounding better at lower current output but must have been mostly the op amps, the OPA1622 was designed for high output current but the current is on its input for I/V.
Last edited:
Whether a larger value feedback R is preferred depends on the circuit configuration. In a 'textbook' I/V stage, there isn't any filtering between DAC and opamp. In such a case, just scaling up the feedback R isn't going to do much good as then the opamp's got to spend more of its time slewing as its gotta slew further. The higher R means higher output voltage swing for a given DAC output current.
If you introduce a transformer then things get a bit more interesting as you can feed the opamp a lower current than the DAC puts out - then a larger R becomes practical and may well result in an audible improvement.
If you introduce a transformer then things get a bit more interesting as you can feed the opamp a lower current than the DAC puts out - then a larger R becomes practical and may well result in an audible improvement.
I've tried that trick too, current source biassing the outputs and running two opamps in anti-phase. Subjectively still not as satisfying as an SE classA discrete circuit which can be made to generate no load-induced ripple, at least to a first order. The difference is primarily in the HF - the classA-biassed opamp sounds clean enough by itself but then when listening to the discrete circuit afterwards, there's a fairly obvious step up in HF cleanliness.
No current source needed. AVCC and Vref do it if the I/V feedback resistor is selected appropriately. The trick with most opamps is use use a suitable mix of film caps on the +-15v rails. Just don't try it for AD797, IME. I use OPA1612, same as ESS uses and AKM use for I/V. They don't know about the film caps though, but even without the caps, OP1612 is still probably the best choice for modern switched resistor sigma delta dacs. If it doesn't seem that way, something else is wrong with the design, IME.
A current source just happens to be the way I chose to arrange the DC bias. Using the feedback R is another choice if there's no downside to having the opamps have DC offsets at their outputs. In my case, there was as I was using a single rail with only just enough swing available - an offset would have sapped too much headroom.
This thread isn't for designs based on switched resistor S-D DACs, rather current source output multibit ones.
This thread isn't for designs based on switched resistor S-D DACs, rather current source output multibit ones.
Right, I was just trying to say the context under which I have found opamps to work fine. The context here is a bit different, so different results may happen. However, the film cap thing may be equally applicable in terms of improving SQ, whether or not it is economically justifiable for a cost conscious project would be another matter.
Context is important yes. It strikes me (do correct me if I've gotten the wrong impression) that you don't have too much choice about your I/V stage, given the numbers for THD+N you're targetting. Its either opampA or opampB as no discrete circuit will come close in terms of measured linearity. Whereas I'm free to play with more options, though I tend to favour the cheaper (i.e. longer in the tooth, with a healthy secondary market) opamps and it could be that more modern ones do indeed sound cleaner in the HF.