That's great, Tom - thanks for that! I am interested because IME behaviour at the highest audio frequencies are where the real weaknesses are; everything starts to get worse, such as PSRR - so, I'm wondering whether the conventional measuring techniques will show anything of interest, at all ...
Hi Tom
Most of the chip amp designs I have read have many pages devoted to tweaking the amp with comparatively expensive high end parts (naked z foil resistors etc) with reported positive results.
I know it's early days yet (and I don't know the BOM components) but do you think that replacing some critical components in your design with an exotic resistor/capacitor or two will improve SQ?
Most of the chip amp designs I have read have many pages devoted to tweaking the amp with comparatively expensive high end parts (naked z foil resistors etc) with reported positive results.
I know it's early days yet (and I don't know the BOM components) but do you think that replacing some critical components in your design with an exotic resistor/capacitor or two will improve SQ?
Last edited:
Most exotic parts are relabeled standard parts. Spending lots of money on them typically makes measurements worse and results in many posts about sighted trials by the person who spent the money. All properly controlled ABX results I'm aware of show subjective preference for better measurement, though some of Harman's results show up 80% of listener preference in speakers can be attributed to other factors. So, you know, if you like the colour of an exotic cap's jacket it might well make you feel better about the amp whilst the Mod's high PSRR renders the supply performance degradation negligible. I doubt Tom would consider this a supported configuration, though.
Resistor tolerance matching on the €6 THAT 1200 is tighter than Vishay's €40 matched Z foil quads. Same is true for the majority of integrated difference amplifiers, whether standalone or included in three op amp instrumentation amplifier topologies like the 1200. There are reasons to use Z foils in certain cases, mainly high precision measurement circuits in non-audio situations where the 5ppm temperature coefficient of Susmu RGs is too large. But equaling or exceeding signal transformer CMRR is not among them.
Resistor tolerance matching on the €6 THAT 1200 is tighter than Vishay's €40 matched Z foil quads. Same is true for the majority of integrated difference amplifiers, whether standalone or included in three op amp instrumentation amplifier topologies like the 1200. There are reasons to use Z foils in certain cases, mainly high precision measurement circuits in non-audio situations where the 5ppm temperature coefficient of Susmu RGs is too large. But equaling or exceeding signal transformer CMRR is not among them.
The MOST important expensive high end part in this design is Tom's PCB and his instructions for how to wire it up.
This WILL improve SQ .. IF you do EXACTLY as he specifies.
For the other Golden Pinnae parts, they only help if they were hand carved from solid Unobtainium by Virgins.
That's exactly right. The performance of the Modulus-86 is largely due to the engineering that went into the board layout. Of course, the circuit plays a role, but, in particular in the midrange and high frequency range (so >1 kHz), mostly the layout determines the performance limit.
Using boutique parts will do an excellent job at draining your bank account. As Twest points out, in many cases the boutique parts will actually provide worse performance than standard parts. Boutique parts are usually also characterized to a much lesser extent than a standard part. With a standard part, at least I can get a data sheet...
There are cases where the type of part matters. This is why I specify film capacitors to be used in the DC servo and the Zobel network and NP0/C0G ceramic capacitors for the compensation networks. It is also why I specify the use of X7R dielectric ceramic caps in the decoupling networks. I also recommend using metal film resistors. All of these parts are available from reputable suppliers such as Mouser, Farnell, Newark, Digikey, RS, et al. All the parts on the Mouser BOM, linked to in the design documentation, follow the above specifications. The total BOM cost is about $45 per board...
~Tom
OK, thanks for your thoughts. I guess there are many others who would disagree 
and there have been a multitude of threads discussing that topic!
I've "upgraded" parts in some of my projects and and IMO there was an improvement and am pretty sure it wasn't because the new part was a nicer color.
Yes, the design of the amp is paramount and if spending more on exotic parts only brings a perceived negligible improvement then why spend the extra dollars? However, I'm not adverse to trying other components (exotic or not) and if I think it sounds better then that's what counts - all part of the fun of DIY'ing, surely.
and there have been a multitude of threads discussing that topic!I've "upgraded" parts in some of my projects and and IMO there was an improvement and am pretty sure it wasn't because the new part was a nicer color
.Yes, the design of the amp is paramount and if spending more on exotic parts only brings a perceived negligible improvement then why spend the extra dollars? However, I'm not adverse to trying other components (exotic or not) and if I think it sounds better then that's what counts - all part of the fun of DIY'ing, surely.
OK, thanks for your thoughts. I guess there are many others who would disagree
Yep. Many disagree. That's fine. I derive no satisfaction from that peeing contest.
The parts do matter. An electrolytic cap in the direct signal path will introduce distortion if not chosen correctly. A polypropylene capacitor will not introduce distortion. A carbon composite resistor will exhibit more 1/f noise than a metal film resistor. Etc. The differences between different brands of resistors or capacitors (with the same specifications) are, however, negligible. The PCB layout will have a much bigger impact on the circuit performance than the components in the circuit, provided that you populate the board according to my BOM.
I suggest that anyone interested in figuring out why many people report positive effects from the implementation of an expensive solution, although, objective measurements report no change or even a performance degradation, to take a look at a introductory text to the human cognition area of psychology. In particular the area of cognitive biases is pretty interesting from that perspective. Matlin, "Cognition", would be my text of choice. For lighter reads, Gilbert, "Stumbling on Happiness", Ariely, "Predictably Irrational", and Kahneman, "Thinking Fast and Slow" are recommended as well.
Illusions are everywhere and in all senses. We humans tend to believe that what we perceive is reality. That's not always the case. Most are familiar with the visual illusions, but similar illusions exist for the other senses.
In the end: Whatever floats your boat... Whatever makes you happy.
~Tom
Last edited:
Yes, thanks for the reading material. I'm now seriously considering buying the Modulus-86 with the part of the funds coming from my proposed "upgrade" of the PS caps for my current chip amp.
I wonder if that means I'm cured
I wonder if that means I'm cured
If you're concerned about subjective SQ (as am I) I'd attack the supplies (meaning lower their impedance) to the THAT chip. It may well be the limiting factor, subjectively as its operating in class B and from the spec sheet there appear to be a number of opamps inside all existing on a rather measly quiescent supply current under 5mA. Hence power supply switching noise looks to be a potential SQ issue.
Yes, thanks for the reading material. I'm now seriously considering buying the Modulus-86 with the part of the funds coming from my proposed "upgrade" of the PS caps for my current chip amp.
I wonder if that means I'm cured
Well, I'm no doctor, but it does appear you now have some knowledge on the subject which allows you to see things a bit more rationally.
The stellar power supply rejection of the Modulus-86 will remove your worries about whether the supply is good enough. I have yet to measure any parameter that was meaningfully different between Mod-86 on a lab supply and Mod-86 on an unregulated, practical supply.
~Tom
As mentioned before, the supply for the opamps on the board, including the THAT1200, is regulated with a local low-noise regulator. The regulator provides 80 dB of line regulation at 120 Hz. The THAT1200 has a PSRR of 82 dB. That's a grand total of 162 dB of supply rejection. Even if the regulator happens to have marginal line rejection, the total supply rejection is still 148 dB. What's your concern again?
I don't see any mention of class B operation in the spec sheet. Would you point me to a page number? I would be very surprised if the opamps in the THAT operated in class B. They're most likely operating in class AB, just as any other opamp.
Please elaborate on how 5 mA for three opamps is "measly" and somehow inadequate for a performance part. The LME49710 uses 5 mA and offers 0.00003 % THD and 55 MHz bandwidth. I do happen to know that getting to 0.00003 % THD required quite a bit of current. The higher bandwidth requires more current as well. The THAT1200 is specified to 0.0005 % THD, 22 MHz, 5 mA. That sounds about right for three opamps at that performance level, actually.
Switching noise?! Please elaborate. Where's the switching supply? That claim is grabbed out of thin air with no foundation in reality.
~Tom
Last edited:
As mentioned before, the supply for the opamps on the board, including the THAT1200, is regulated with a local low-noise regulator. The regulator provides 80 dB of line regulation at 120 Hz. The THAT1200 has a PSRR of 82 dB. That's a grand total of 162 dB of supply rejection. Even if the regulator happens to have marginal line rejection, the total supply rejection is still 148 dB. What's your concern again?
Load regulation, not line regulation
I inferred it, so no page number is to be found. Have you met any opamp with a classA output stage?
OK then, class AB. Same argument applies as for classB in regards to the power supply noise.
You will need to clarify your question to get an elaboration from me. 'Performance' is determined subjectively, or via measurement? In the DS (title page) I counted four opamps, not three. Page reference for three please?
The switching noise I was referring to was haversine-shaped current pulses flowing in the classAB output stage. Any clearer now?
Some you can drive into class A by applying an external current. All opamps I've seen have been class AB (not B).
Recap of amplifier classes (Wikipedia).
In the Modulus-86, the THAT sees a high-impedance input. It delivers the input bias current for the LME49710, so nA... The THAT never leaves class A in the Modulus-86.
Recall, as long as the load current is less than the quiescent current in a class AB output stage, the output stage operates in class A.
My bad. The THAT1200 contains three buffers and one opamp. A buffer does not have to be an opamp. It can be as simple as an emitter/source follower.
~Tom
I take your point, I was using 'classB' as a place holder for 'not classA'. Driving the internal opamps (or buffers) into classA isn't an option here, only the last one can be so biassed. The most heavily loaded one (4kohm effective) is internal.
How can you be so sure? Have you considered the internal loading on the opamps (resistor values are given in the graphic on the first page)? Have you also knowledge of the standing output bias currents in the design? If so do please share.
My concern though is for the power supply noise, which can still be present even with a classA push-pull output stage. Only a CCS loaded, shunted SE OPS would be immune to this.
Yet the graphic does label them all 'OA<n>' which does to me suggest they're opamps. An emitter follower would not give DC precision (10mV max output offset spec notice) so seems rather unlikely.
I don't have any inside information for the THAT1200. If I did, I would be working for THAT and not be allowed to give it out. However, as an IC design engineer, I have designed several opamps, so I have a good general idea of the orders of magnitude for these things. I would be extremely surprised if the output stage of the opamps/buffers in the THAT1200 is biased at less than a few hundred µA. The 5 mA supply current should be a clue for you...
Even if the output stage is lightly biased at, say, 100 µA, that's still over four orders of magnitude (10000x) above the typical input bias current of the LME49710. It would also be well over three orders of magnitude (1000x) above the worst case input bias current. In addition, the THAT drives the 30 kΩ differential input impedance of the LME49710, so 90 µA (peak) of signal there. Even at very light bias, the THAT is not likely to ever enter class AB.
Anybody curious about my background can find my resume/CV on my website.
The regulator has an output impedance in the range of 1-10 mΩ. The 90 µA signal current from the THAT1200 will result in 0.9 µV of ripple on the supply. This is attenuated by 82 dB by the THAT1200, so there'll be 90 pV of supply related crap on the output of the THAT. This will contribute about 0.9 nV on the output of the MOD86. That's 56 dB below the -125 dBV noise floor. Does this really keep you up at night?
No it wouldn't. An ideal CCS loaded ... will be immune. A real CCS loaded ... will not. If you don't believe me, I suggest that you set up a simulation. Include the output impedance of the CCS and watch what happens.
Depends. There are ways around that. Also, don't assume that the equivalent circuit schematic tells the whole story.
~Tom
Indeed, I was going on the assumption it would be around 200-300uA. So where's the answer to my question about the internal loading? Seems you've ignored that totally and its the fly in the ointment of your (apparently) faith-based claim that the THAT never leaves classA.
True, but irrelevant here. See above.
No, because its entirely beside the point. See above.
Where have I made that assumption, please?
I apologize for being human. It's getting late on this side of the pond.
I'll take your 4 kΩ load at face value. That's still only around 800 µA peak output current for the THAT (including the 90 µA that goes into the LME49710). Even if the THAT was operating in Class B (not AB), that would still only lead to 8 µV of bounce on the supply. That would result in 0.635 nV on the THAT and 6.35 nV on the output of the MOD86. That's -164 dBV or 39 dB below the ultra low -125 dBV noise floor.
There's no issue there. None.
~Tom
Yeah, and since it's probably mostly or all class A there's less actual variation in load current. So it'd be closer -infinity dB, actually. Time for a new windmill for the thread to tilt at, I think...
Well, if you push a polypro to really crappy operating point the THD might get above -125dB. In the data I've come across C0G, PPS, PET, PEN, and audio signal path electrolytics like Elna RFS all seem to measure in the -140 to -130dB range in typical audio operating conditions. It is possible to touch that level of linearity and dynamic range with an audio op amp in limited circumstances but my experience is loop gain and noise floors in practical circuits are such that performance is often 20+dB worse than this. So it's reasonable to treat well chosen caps as ideal most of the time.
I'd be interested in the ABX results if you have a link. Whilst there are pages upon pages nearly all of them are sighted "tests" without much, if anything, in the way of controls.
In this kinder and gentle age we live in it's also important the unobtainium is free trade and the virgins volcano and Huitzilopochtli free.
Last edited: