I/V opamps: selecting/tweaking feedback caps + resistors

[Calvin]

Hi,

the PCM1794A puts out a center current around which value the signal current swings. The resistors R1 and R2 convert this current into a voltage after the formula U=R*I. At the output of U1 and U2 a dc-voltage appears onto which the signal voltage is added. Since U3 is connected as a differential to single ended amplifier, the dc-part of the voltage cancels while the two out-of-phase-signal-voltages add up.
At the output of U3 only the signal voltage is left.
C1 and C2 need to be designed such that they attenuate the image frequencies of the DAC itself (which depends on the frequency the DAC is clocked with, say 96kHz or 192kHz). Typically this gives lower values than those suggested in this application sheet. The problem is, that the OPamps are simply too slow in their response. The PCM1794A´s output settles within 200ns to a new value. The circuit connected to the DACs output must be at least as quick (see settling time values in the datasheets) as the DAC itself, or the signal becomes seriously distorted.

In this case however, the caps are enlarged in value so that the DAC´s output-current needs more time to charge the caps. The DAC is deliberately made so slow that the OP can keep up with its speed.

To my experience it is sonically advantageous to clock the DAC as high as possible and to omit with as much filtering as possible, or at least to reduce filter steepness as much as possible. This asks for high speed circuits, wether they be made of OPamps or discrete circuits, with high slewrate values and small settling time values. Examples are OPA1632, THS4031, THS4021, THS4150, AD4932 et al.

TI´s application notes give a more detailed insight.
Application notes:
- BB/TI, SBAA150-April2008 "Design for a wideband, differential transimpedance DAC-output"
- National Semi, AN-1083 "Design considerations for a transimpedance Amplifier"
- BB/TI, SBOA122, Nov2009 "Transimpedance considerations for High-Speed-Amplifiers"
- BB/TI, SLYT360 "Interfacing op amps to high-speed DACs, Part2: current-sourcing DACs" (The PCM1794A is a sourcing DAC, SLYT342, Part1 deals with currcent-sinking DACs)
Remark: In contrast to some of the DACs mentioned in the Apps, the PCM1794A likes to see a virtual ground (low impedance current path) at its outputs and the common-mode voltage is 0V. The voltage appearing at the DAC´s output must be kept small, since the outputs are protected by diodes which start conducting (hence distort) at voltage levels >100mV.

jauu
Calvin

 

[abraxalito]

Here's the MFA324 output schematic...

OK, I can see all manner of problems with that output stage. Are you sure you'd like to rescue it? - sometimes a clean slate is better If I was wanting to build a beefy output stage similar to this one I'd go direct to the LM1876 - a nice, low component count solution.

If you'd like to play around with this discrete design, then adding some resistors and changing the transistors would be a first step. The MPSA43/93 is a poor choice here - its a 200V part and has low beta (min 25) at low (<1mA) collector currents from what I can find in the datasheet. An hFE vs IC plot would be nice, but I haven't found one that looks reliable. BC547B/BC557B would be a step in the right direction but I'm sure there are plenty of others.

As regards the extra resistors I mentioned, these need to be added to degenerate the input pair and the current mirror load. This will improve linearity which is a fairly major issue here as the DAC's putting out shed-loads of HF noise. With degeneration the input pair can be run at a higher tail current without compromising stability. While on the subject of stability, I haven't run a SPICE simulation but my gut tells me this circuit isn't going to be highly stable as its running unity gain and only has 47pF compensation cap. Whereas similar circuits in Doug Self's comprehensive book on power amps have 26dB gain.

So, to details - the emitter resistors for TR101,102 are 100R, same value in the emitters of TR104,105 to linearise these. Increase the tail current by reducing R109 to 1k1.

Next up - the output darlingtons are horrid. And they look to be underbiassed too as there are only 3Vbes plus 500mV (R116). So heaps of crossover distortion potential there. At such low IC's the current gain of these devices is falling off substantially (already <600 @ 100mA) and they'll be running nowhere near 100mA here. Would you like to measure the voltage drop across R114 and 115 in quiescent? I'll do a little research for suggested replacements for these but by now you might be getting the picture that the LM1876 is the path of least resistance...

 

[hollowman]

MFA324 upgrades

OK, I can see all manner of problems with that output stage. Are you sure you'd like to rescue it? - sometimes a clean slate is better If I was wanting to build a beefy output stage similar to this one I'd go direct to the LM1876 - a nice, low component count solution.

Hmmm so you're suggesting a chip amp to replace its stock output (which is, TTBOMK, "beefy" to begin with)? Won't the LM1876 output be a bit too much ... I mean for just a D/A processor (line level) output topology? Don't have much experience here so I may be behind the times a bit!

If you'd like to play around with this discrete design, then adding some resistors and changing the transistors would be a first step. The MPSA43/93 is a poor choice here - its a 200V part and has low beta (min 25) at low (<1mA) collector currents from what I can find in the datasheet. An hFE vs IC plot would be nice, but I haven't found one that looks reliable. BC547B/BC557B would be a step in the right direction but I'm sure there are plenty of others.

As far as trans. choice ... you're right: those MPSA's are probab not optimal. Dunno why Musical Fidelity used them **? I've heard good things about Zetex 651/751, which may be a step up from the BC547B/BC557Bs. Also see possible potential (??) in replacing the 4148 diodes with LEDs?

** The MF A324 is a stange unit. Not sure the folks at MF are all upstairs if you know what I mean. Cases in point: The chassis (super heavy-gauge metal; thick panels) and parts of power supply are overkill. Also, it has (uh ... had, before I took it out, that is!) an SPDIF output why waste $$ on such a useless features?.
The A324 was Class-A rated by Stereophile mag from roughly 2002-2004. I think it was the cheapest Class-A-rated D/A they ever "promoted" (lot of internal politics and promotion here I'm sure!).

 

[hollowman]

Meant to add to my last comment (before edit time ran out!):
GerorgeHiFi dramatically improved SQ, and I made other simple mods ... all leading to way better sound than the stock "Class-A" unit! Hmmm...does this mean what I have now is in the same league as $15K dCS's?

 

[Rick Miller]

I like the Analog Devices AD811 for I/V. Walt Jung wrote about it in the Audio Amateur.
I put a 100uh inductor in front of it as a pre filter.

 

[jcx]

Better op amps, Better topology for IV?

I don't know if anyone has actually built one but I'm having some success simulating Hawksford's multiple nested loop op amp I/V

http://www.diyaudio.com/forums/digi...sford-iv-nested-loop-op-amps.html#post2218591

I show much better op amps than the original Hawksford IV paper circuit


there are several active I/V threads in the Digital Source sub forum with various circuit and modern op amp selection discussion

 

[hollowman]

I don't know if anyone has actually built one but I'm having some success simulating Hawksford's multiple nested loop op amp I/V

http://www.diyaudio.com/forums/digi...sford-iv-nested-loop-op-amps.html#post2218591

I show much better op amps than the original Hawksford IV paper circuit

Very cool. But so many choices now....

there are several active I/V threads in the Digital Source sub forum with various circuit and modern op amp selection discussion

Okay, so given all these choices which [opamp I/V and/or complete opamp-based output section] may be considered optimal pair-ups with:

Modern TI/BB DACs (PCM1792, 1794, other 179x, etc)
Modern AD DACs (AD1853, AD1955, etc.)
Modern Cirrus DACs (CS4398)

Or can most (or better of the lot) be considered "universal"?

Finally, if one is upsampling prior to the DAC, then whihc opamp-base I/V and/or complete output section is "optimal"?

 

[abraxalito]

Hmmm so you're suggesting a chip amp to replace its stock output (which is, TTBOMK, "beefy" to begin with)? Won't the LM1876 output be a bit too much ... I mean for just a D/A processor (line level) output topology?

Can't really follow you here - 'too much' what? Too much level? - but those discrete darlingtons can put out too much level just as easily. However the output level is determined by the gain and that'd stay the same with the chipamp - unity gain.

Don't have much experience here so I may be behind the times a bit!As far as trans. choice ... you're right: those MPSA's are probab not optimal. Dunno why Musical Fidelity used them **?

Probably they're using them elsewhere in amps so they had them in stock. Adding new parts to inventories can get to be expensive.

I've heard good things about Zetex 651/751, which may be a step up from the BC547B/BC557Bs.

Certainly a step up in price I'd imagined you were looking for the best bang for the buck so to speak. I've used the ZTX - they're really nice parts but quite unnecessary to go to such expense here. Those come into their own where high dissipation is needed in a small package.

Also see possible potential (??) in replacing the 4148 diodes with LEDs?

Can't for the life of me see any possible benefit there except that an LED would have a greater forward voltage than two diodes so may well improve the crossover distortion. But you'd need to check that the bias hadn't got out of control - simply changing the two diodes for one LED might cook the darlingtons (I have no idea if they're heatsunk).

** The MF A324 is a stange unit. Not sure the folks at MF are all upstairs if you know what I mean. Cases in point: The chassis (super heavy-gauge metal; thick panels) and parts of power supply are overkill. Also, it has (uh ... had, before I took it out, that is!) an SPDIF output why waste $$ on such a useless features?.

Don't get me started on the unreasonableness of much hifi engineering...

The A324 was Class-A rated by Stereophile mag from roughly 2002-2004. I think it was the cheapest Class-A-rated D/A they ever "promoted" (lot of internal politics and promotion here I'm sure!).

... or on the politics of hifi magazines and reviews.

 

[abraxalito]

I put a 100uh inductor in front of it as a pre filter.

What kind of inductor - ferrite or air cored for example?

 

[hollowman]

Certainly a step up in price I'd imagined you were looking for the best bang for the buck so to speak. I've used the ZTX - they're really nice parts but quite unnecessary to go to such expense here. Those come into their own where high dissipation is needed in a small package.

Hmm...abraxalito/anyone: other than ZTX, what other price-no-object brand/model replacement trans are there for this kind of app. Put another way, what is the Black Gate or Mundorf of this size/type of trans?

 

[hollowman]

Chipamp LM1876 for output stage of D/A processor

Can't really follow you here - 'too much' what? Too much level? - but those discrete darlingtons can put out too much level just as easily. However the output level is determined by the gain and that'd stay the same with the chipamp - unity gain.

So can one "just" throw in the LM1876 after the I/V opamps? Also, would a LM3886 be better choice? Or how about other manuf's of chipamps (if any). (Sorry, dude: no experience with chipamps on my end!)

 

[abraxalito]

So can one "just" throw in the LM1876 after the I/V opamps?

No, the LM1876 would need a couple of RC networks to keep it stable at unity gain as its not compensated for that. But that's fairly easy. Otherwise, yes just drop it in.

Also, would a LM3886 be better choice?

Well obviously not or I'd have suggested it first, duh! The LM3886 has higher current output and is only single channel - we don't need its high current here and we do want two channels. So in my estimation, the LM1876 is better suited, and I have experience in using it as a line output so I know its distortion is very low into easy loads. I've only used the LM3886 as a power amp.

Or how about other manuf's of chipamps (if any). (Sorry, dude: no experience with chipamps on my end!)

Sure, the LM1876 isn't the lowest cost route - here, if you don't mind using two chips then the TDA2030 is a really cheap-skate choice and I believe its not half bad sonically. But no experience with it as a line level driver so just a smidgen more risky.

 

[hollowman]

No, the LM1876 would need a couple of RC networks to keep it stable at unity gain as its not compensated for that. But that's fairly easy. Otherwise, yes just drop it in.

Was a version with RC networks ever posted on this forum (or other sites)?

The LM3886 has higher current output and is only single channel - we don't need its high current here and we do want two channels. So in my estimation, the LM1876 is better suited, and I have experience in using it as a line output so I know its distortion is very low into easy loads. I've only used the LM3886 as a power amp.

I guess I'm having issues seeing the big picture(s). (and this may be entirely due to my lack of understanding/experience w/chipamps, & notably using them in line-level apps)....
First: TTBOMK, the LM1876 is meant to drive speakers, so its output impedance may not be ideal (esp. dynamically).
Second: how does one "safely" drive a following preamp or inputs to a headphone amp with something (LM1876) that could (at peaks) be outputting 20W!?

 

[abraxalito]

Was a version with RC networks ever posted on this forum (or other sites)?

Maybe not with the LM1876 but ThorstenL I think may have posted one with another chip amp.

I guess I'm having issues seeing the big picture(s). (and this may be entirely due to my lack of understanding/experience w/chipamps, & notably using them in line-level apps)....
First: TTBOMK, the LM1876 is meant to drive speakers

True enough but then the darlingtons in the circuit you're replacing could too (admittedly minus the 10R emitter resistors). I'm just offering something roughly equivalent to that. Really serious high end (solid state of course) stuff often has mega-beefy output buffers, so this trumps those rather expensive pieces of kit with BUF634s, LT1010s and the like. It also means if you like headphones, you can drive them direct (probably through an attenuator in some cases).

...so its output impedance may not be ideal (esp. dynamically).

The output impedance can be bumped up with a series resistor - I'd suggest 10R then it'll be unconditionally stable. In fact output impedance isn't really something to be concerned about for a chip amp, the much higher current limit over a normal opamp might be.

Second: how does one "safely" drive a following preamp or inputs to a headphone amp with something (LM1876) that could (at peaks) be outputting 20W!?

The LM1876'll only put out 20W if you give it an 8R load or lower and even then, only when driven to the rails with higher supplies than you've got here. I bet your preamp doesn't have such low input impedance. And remember its output voltage is limited to considerably lower than the supplies because of having only unity gain after the I/V converters.

 

[jcx]

if the discussion is about a Line Level output for a DAC then the discrete Lin with monolithic Darlingtons is a bad joke

look at Walt Jung's multiloop line drivers - I'd use TPA6120 instead of AD811 but either is faster than the output Q of the mfa324 circuit

 

[Rick Miller]

What kind of inductor - ferrite or air cored for example?

I use a single layer ferrite core inductor. I would prefer an air core, but the size of the coil, and resistance of the wire would cause problems. In this application the ferrite does not seem to be audible, and this inductance before the op amp really helps limit the RFI going into the input of the op amp.

 

[hollowman]

Speaking of chipamps, what about Tripath TA2020 for the above D/A app.?

 

[garfieldso]

I have done some mods to my Dr.DAC2, and also replaced the PCM1798 with PCM1794; after a lot of testing on I/V opamp With LME49720, OPA627AP..etc .

So happen found that the best was using the TI's NE5534 as what the datasheet recommended. Also discovered that 2xNE5532 is totally not equal to 4xNE5534 in sound performance. Sound stage of NE5534 is broader and deeper, freq response is in equivalent balance smoothly.

Another thing is that when testing the opamps, at least let the new one to has burn-in enough prior jump into comparison conclusion, otherqise you may throw away a lot of good things.

 

[hollowman]

Opamp burn in time -- NOT

Another thing is that when testing the opamps, at least let the new one to has burn-in enough prior jump into comparison conclusion, otherqise you may throw away a lot of good things.

IME, the opamp itself requires zero "burn-in/break-in" time. A lot of DIYers modding w/ and/or rolling opamps also (concurrently) replace associated caps, the electrolytic variety of which do need some BI time (e.g. Pana FM).
Ahh ... but who has time to replace all these components independently to test out BI times ... so we end up with claims of "opamp BI".