Designing a universal diff-in/diff-out Head Amp

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
There are quite a number of circuit diagrams for (differential) Head Amps available;
point is that most of them are either very complex, consuming large amounts of energy, or to be assembled from a large number of selected Fets as input devices.
To make a difference, I used the following points:

Diff in
Diff out or SE out
No caps in the signal path
To be either supplied by a 5V power bank or a USB mains supply.
As simple, yet as powerful as possible.
Selectable as a voltage or as a current input circuit.
Wide range of selectable Gain settings


Latest Update

I noticed that the dropbox link with all circuit diagrams, pictures, Boms and Gerber files was no longer active,
that's why I reinserted it here:
Dropbox - Zip Files - Simplify your life

And here is a link with all the specifications:
Dropbox - This universal Diff In.pdf - Simplify your life


Hans
 
Last edited:
The difference between a current input and a voltage input diff amp, is rather small, as can be seen below.

Edit: Image is posted on #6

With a handful of components, both can be realised in one go. There are two inputs visible. When connecting the floating Cart to input 1 and a short circuit to input 2, you will have a current input amp. R1 and R2, in combination with the Cart’s internal resistance will determine the gain. Connecting resistor R6 to input 1 and the Cart to input 2, will result in a voltage input amp, where the ratio 1+(R1+R2)/R6 will set the Gain and R3 + R4 serving as the Cart termination.

As mentioned in point 6), for a 0.5mV Cart, U1 and U2 can be the 0.9nV/rtHz AD797 and you are ready to go with a minimum amount of components, yet hitting the target of 75dBA S/N with a 20dB Gain setting with a THD below -90dB up to 20 Khz.

For a 0.25mV Cart, you will need amp’s with halve the noise as produced by the AD797. For this I used John Curl’s patent, dating back more than 40 years ago, but now modernised with components that weren’t available at that time. Basically the circuit looks like:
 
Last edited:
The circuit is fed from a double 1.5 Volt supply. John Curl used 8 paralleled transistors in his design, but with modern ZTX851/ZTX951, two transistors are adequate to meet the noise target of 0.64nV/rtHz after Riaa and A-weighting. In said Patent, the circuit was used in a current input setting, with a resistor directly between in and output. But the circuit can just as well be used in a voltage input mode.

There are a few flaws to be mastered however, being uncontrolled current to flow into the Cart and the far from zero output impedance. When powered by batteries, but also when powered from a mains supply, the slightest difference in V+ and V- or a difference in Hfe or Vbe between both transistors will cause a current to flow into the Cart. It might be harmless as some say, but it is better to stay on the safe side. This can be solved with an LT1884, having only 50uV input offset, regulating the positive supply with a Fet. This is a very simple and effective mechanism, keeping the Cart current below 1uA at all time.

Second point is the high output impedanceof the circuit that heavily affects gain of the circuit when loaded by some feedback resistor. To overcome this, I have added a LT6203 op-amp as buffer, a 1.5 nV/rtHz videoamp that can operate from the same supply, thereby giving a low output impedance.

In a differential setting, meaning doubling the amount of all components, total current is ca. 17mA. When using 1.5 Volt Alkaline penlight’s, you could power this circuit for almost 90 hours. With 1.5 Volt Alkaline D cell’s even 900 hours !
 

Attachments

  • Curl4.jpg
    Curl4.jpg
    141.7 KB · Views: 2,135
The complete sub-circuit in its most complex form with these added features is shown below. One could however leave the LT6203, only making it a bit more complex to find the right value for a feedback resistor but with hardly any effect on THD. So in its most basic form, only U1 is stronglyadvised !

Both the LT6203 and the LT1884 are duals,so only two SOIC packages have to be used, adding almost nothing to the neededPCB space.

There might be some concern that at powerup some uncontrolled current could flow into the Cart. However, because of thedifferential set up with no connection to Gnd, this is avoided and needs nofurther attention since being far under 100uA Imagine having a 10 Ohm Cart, that is used in current mode. 0 dB@10Khz, means ca. 5mV rms. In that case+/- 700uA peak flows through the Cart. So one can safely assume that start-upcurrents below 100uA are harmless.
 

Attachments

  • Curl5.jpg
    Curl5.jpg
    232.4 KB · Views: 2,167
Now the proof of the pudding. Four configs for resp. current / voltage / 0.25 mVolt and 0.5 mVolt all put in a Simulation.

To start with, a 4nv/rtHz differential input MM was configured with a 580mH/460 Ohm Cart. This was done by means of two input buffers with parametrised op-amps set to the exact noise power.

The 0.25 mVolt current input version had lower noise as its voltage brother, just as expected. But for a reason that is unclear to me, theAD797 produced more noise in the current mode. This current mode version however performed just 0.25dBA below the 75dBA target, an insignificant and still acceptable difference.

All versions produced a very low THD, below120dB from 20Hz to 20Khz at 0 dB. CMRR was way below 180dB up to 20Khz, because of the perfect match of allresistors. In practice one can easily keep this under 80dB with 0.01% resistors at vital positions. To put this in perspective, one can touch the Cart’s wires with bare fingers, without hearing anything at all coming from your speakers.

For the simulation results, see images below.
 

Attachments

  • Curl6.jpg
    Curl6.jpg
    741.2 KB · Views: 2,067
  • Curl7.jpg
    Curl7.jpg
    617.7 KB · Views: 1,973
Member
Joined 2014
Paid Member
Hi Hans,

Couple of questions. Have you considered the case of digitising case where you are going flat into the ADC and only need around 20dB for MM and 40dB for 0.5mV MC?

Also regarding CMRR. You don't gain anything until you actually do a conversion. One idea I have seen is this http://www.thatcorp.com/datashts/dn140.pdf#page=14 . Contentious outside of pro circles but interesting to consider.
 
Hi Bill,

I concentrated mainly on the function of a Head Amp, being to bring its input voltage to the level of an MM Cart. However, there is absolutely no problem to give the 0.5mV version, using the AD797's, a gain of 40dB. Just a matter of changing 2 resistors. But it would also be easy to make a second diff-in/SE-out 20dB gain flat amp suited for MM Cart's, for instance with an OPA1642 a 1652 and even a 1632.

Maybe you can go in a bit more detail regarding CMRR. If I understand you correctly, your point is that CMRR is dependant on the whole chain involved, including the diff to SE conversion, a point that I fully support. But I don't get the message so much how this relates to page 14 in your link in this very case. What ADC do you have mind, does it expect a diff input signal and does it support CMRR, or do you have to remove the CM signal before offering it to the ADC ? It's all possible, op amps to perform the needed functions are all on the shelves.

Hans
 
Member
Joined 2014
Paid Member
Hi Hans,

No problem, as it's something I still worry about and may have missed the point. If you are straight balanced all the way through, then overall CMRR can be destroyed by one poorly selected resistor*. The Cohen cross coupled circuit, whilst originally designed for receiver purposes can be used as the output drive stage, removing CM signal so you effectively get a clean balanced and differential output.

If this matters other than paranoia is of course a different issue :)

*There is the Whitlock INgenius designs for that of course.
 
Bill,

I have the feeling that you are overly worried for CMRR problems. Just to respond to your question, I have modified the MM preamp into a real world flat 20dB MM preamp, with the OPA1652. This circuit still has the 4nV/rtHz equivalent input noise after Riaa and A-weighting with a 580mH + 460 Ohm Cart.

The input circuit amplifies 40dB, and has an excellent CMRR of 40dB insensitive to resistor values deviating a few percent from nominal. See R3 vs R22, one has the correct 4k95 value needed for a gain of 40dB and the other is 4k5, 10% off.

This input circuit is followed by a diff to SE convertor, attenuating -20 dB. This circuit is very sensitive to a correct resistor value. But as you can see I made R17 1K1 instead of 1K, again 10% off the ideal value. Therefore this convertor only has 20dB CMRR. However, in combination with the input circuit, CMRR becomes 60 dB, see image below. When using a 1% instead of a 10% resistor, you already get 80dB CMRR for this MM circuit alone. Measuring CMRR from the Cart upwards, adds another 20dB because of the Gain of the Head Amp.

So with only 1% resistors a 100dB CMRR, how hard can it be ?

Hans
 

Attachments

  • Curl8.jpg
    Curl8.jpg
    417.5 KB · Views: 836
Hans have you considered phantom power? Let's say for LP transfer using a popular external device like the Focusrite 2i2 which has phantom power for mics and decent low noise preamps. My results have been very encouraging with software RIAA. There are a lot of folks that want a simple solution to transfer LP's, and I like the idea of no batteries no external supply. etc.

I've been a little hesitant to promote this heavily because I actually think there is a nice cottage product in there somewhere.
 
Last edited:
Member
Joined 2014
Paid Member
I run all low level off silentswitcher, but as I'm considering a 2i2 would be interested in if that is workable. The thing that annoys me a bit about the 2i2 is tracking of the gain pots but I am sure it could be POOGEd.

Hans: Your last diagram shows what I would consider as impossibly high CMRR. 90dB is normally considered the best that solid state can do and transformers are up to 115dB at low frequency. It doesn't feel right, but I can't immediately work out what it wrong.
 
Hi Bill,
I don’t know where this wisdom comes from, but such a general statement cannot be made. In case of my measured CMRR it was the CMRR of the AD797 that was limiting this figure. Look in the specs of this opamp a see how close it follows my measurements. Had this been higher, I would without a doubt have measured higher figures.
Hans
 
Hans: Your last diagram shows what I would consider as impossibly high CMRR. 90dB is normally considered the best that solid state can do and transformers are up to 115dB at low frequency. It doesn't feel right, but I can't immediately work out what it wrong.

Bill you are probably thinking CMRR of a whole system, cable runs, source imbalance, etc. The raw CMRR directly at the pins of a bi-polar op-amp can easily be 120dB or more.
 
Member
Joined 2014
Paid Member
I was, but because Hans was also claiming 100dB CMRR with 1% resistors by dint of cascading CMRR rather than another technique. I remain unconvinced till I have my head around it, esp in a balanced configuration as my (mis?)understanding was based around CMRR being normally measured balanced to single ended, not balanced to balanced. I'll get there.
 
Bill,
Let me try to explain a bit in more detail. That may help to get more insight into the matter of CMRR in this case.

Please go back to the 20dB gain MM preamp that I presented in #10. This Amp starts Diff and ends SE, that's how I measured, just according to your expectation, but how is this signal composed?

The first stage with U8/U9 has a diff gain of 100, but has a CM gain of 1. The CM signal is still there but with the same amplitude as offered at the input. This CM signal can and will be further attenuated by the second stage with U6.

There is however a second mechanism at work in the first stage. When offering a CM signal to U8/U9, the + and - inputs of both OpAmp’s, in this case the OPA1652, will only see the same signal in an ideal situation. However in reality the CMRR of this OpAmp goes from ca. 125dB at 100Hz to ca. 90dB at 20Khz, see image below, but the two OpAmp’s will always be a bit different in their CMRR. This means that between their outputs, a conversion takes place from CM into DM which will be added to the CM mode signal. This DM mode part can never be removed by later stages and will be simply processed as every other DM signal.

So to put things in practice, suppose we use a 100mV input signal and suppose one opa1652 has a CMRR of 100dB at 1Khz and for the other 110dB. Now in CM, will this result in two 100mV CM signals coming from U8/U9, but with a slight difference of +/- 1uV at output U8 and +/- 0.3uV at output U9, both having the same polarity, resulting in a 0.7uV DM signal.

So in effect we now have a 100mV CM signal with a 0.7uV DM signal on top. Now entering U6, the DM to CM convertor, having one resistor 1% off the correct value, CMRR will therefore be 40dB and DM gain is -20dB. So the CM signal from U8/U9 will be attenuated by 40dB from 100mV into 1mV SE. The 0.7uV DM signal will be attenuated by U6 by 20dB into 0.07uV SE, and will completely vanish into the other 1mV. CMRR in total is now 20*log((100mV*10)/1mV) = 60db, just as presented in #10.

However when further fine-tuning CMRR of U6 to a vanishing low value, there comes apoint where the 0.7uV starts to become a border that cannot be passed. In that case CMRR will be 20*log(100mV*10)/0.07uV)= 143dB@1Khz.

Would I have used 20dB gain for U8/U9 and 0dB gain for U6, CMRR with the 1% off resistor would have been 40dB instead of 60dB and the absolute limit would have been at 123dB@1Khz.

Hans
 

Attachments

  • Curl10.jpg
    Curl10.jpg
    106 KB · Views: 597
Member
Joined 2014
Paid Member
Hi Hans,

I'd been processing this as a background task whilst dealing with the matrix this morning (GUI feh, real men use VT100) and the penny had dropped as to what I had been missing. I think the pain of matching everything as close as I could for the balanced Aurak build had warped me a bit* and I wasn't thinking straight and I realised that it's the diff to SE conversion in the first stage that causes the problem in pro audio setups.

Has also given me a pile of other stuff to question over CM impedance stuff that I clearly haven't got my head fully around yet*.

It's a shame but never could find the words to explain to AndrewT properly about why a diff input has a CM gain of 1 and it's too late now...

*This warping of course I blame you for as you stated 0.01% matching. I did however find interesting numbers in the distributions of components. Experience is never wasted.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.