Universal buffer/headamp based on OPA1622

I’ve been following the thread "New Audio Op Amp - OPA1622" for a while and at some point decided to make myself a new universal buffer/headamp based on the OPA1622. The idea is to use it as an output buffer for my standalone volume controller/source selector with the direct headphones drive capability. Since I already had the low noise PSU board design based on the TPS7A3001/TPS7A4901 complementary regulators I just added the buffer part to it and it seems to look good together. I also added the transformer CRC snubber to the board following the guidance in the "Simple, no-math transformer snubber using Quasimodo test-jig" thread. According to the datasheets the LDO regs which I used in the PSU should be able to drive the OPA1622 opamp to near its maximum current output. I guess I do not even need additional gain and can just configure the OPA1622 as a unity gain buffer or have a gain of 2 at most for 32 Ohm headphones which I am using.

It will be a bit challenging to solder those tiny IC packages and 0603 resistors and capacitors around the opamp but I’ve done it already on the other builds and it turned out successful after a bit of practice.

The resulting PCB design and schematic are attached. The PCB is only 3x8 cm and the mounting holes are on the 10 mm grid.

Comments and suggestions are welcome.

Regards,
Oleg
 

Attachments

  • OPA1622.HeadAmp.Sch.jpg
    OPA1622.HeadAmp.Sch.jpg
    566 KB · Views: 4,024
  • OPA1622.HeadAmp.PCB.png
    OPA1622.HeadAmp.PCB.png
    139.4 KB · Views: 3,873

Shaq888

Member
2009-02-09 7:23 am
Hi

Can say that 1622 is quite nifty opamp. Have a look how it drives 1kHz, 10kHz, 20kHz and 50kHz 1Vrms (0dBVrms) into 16Ohm load. Have more loads/voltages if you are interested:
4H4FIot.png

GH1PFwU.png

VFMooAV.png

CmSUJzl.png


Now a few words about you design.
If you are driving heavy load, you need large bulk capacitors next to the opamp. By the "large" I mean a few (or more, if space allows) caps 6.3-8mm diameter, low ESR and value as big as you can afford in this size. Regulators are not as important in this case. I would probably relapse them with simple filters on one transistor, or would not place anything at all.
You output needs an adequate filtering/damping. Headphone cable looks like capacitive load on low-ish freq and like a poorly terminated transmission line on higher freq. Without it opamp like this may have problem with stability. The are some cases when it is barely required, but you have to verify that it is OK this way first, even though it claims "Stability with >1nF cap loads in a G = -1 configuration, no isolation resistor"
You need at least basic EMI filter on input. This opamp has non-degenerated BJT pair on input. It will be prone to parasitic RF demodulation.
Having your input and output terminals apart for left and right is sub-optimal and unlikely practical, since you source and load will have a common ground. Have a thought about joining L and R together.
Unlikely you will need a lot of gain if your source is something like 2Vrms.
Inverting topology is better if you can not match impedances next to inverting and non-inverting inputs, and your mismatch is more than ~100-200 Ohms (roughly)
Two diode bridges are redundant. There is barely a case where it makes a difference, and this is not the one. Put just one.
 
Last edited:
Hi Sergei,

Thanks for the info and suggestions!

If you are driving heavy load, you need large bulk capacitors next to the opamp.

I remember the discussion in the Super Regulator thread which also has remote sensing, where Jan Didden explained that placing large electrolytics at the load defeats the advantages of using the regulators by isolating the load from the regulator. In my understanding regulators "replace" bulk capacitance if they placed close to the point of load. The remote sense (feedback) should take care of lowering the power supply output impedance seen by the load in the audio frequency range or am I wrong here?

You need at least basic EMI filter on input. This opamp has non-degenerated BJT pair on input. It will be prone to parasitic RF demodulation.

I have EMI filter on the preceding stages but I can put it on this board too. I guess you mean small value capacitor to ground between the input resistor and the opamp's non-inverting input to make a low pass RC filter with the corner frequency of around 200~300kHz? Or I need something more sophisticated?

You output needs an adequate filtering/damping. Headphone cable looks like capacitive load on low-ish freq and like a poorly terminated transmission line on higher freq. Without it opamp like this may have problem with stability. The are some cases when it is barely required, but you have to verify that it is OK this way first, even though it claims "Stability with >1nF cap loads in a G = -1 configuration, no isolation resistor"

The absence of load capacitance isolation resistor is what was one of the advantages of using the OPA1622 as I thought. If I still have to use it, then it would make it similar to the other more diy friendly headamp designs, e.g. LME49990/LME49600 combo.

Having your input and output terminals apart for left and right is sub-optimal and unlikely practical, since you source and load will have a common ground. Have a thought about joining L and R together.

Separation of the L and R inputs is happening anyways due to the use of chassis mounted RCA connectors. Outputs on the other hand are relatively close together but I'll probably make a provision for common ground connection by rerouting a bit the EN pin input.

Inverting topology is better if you can not match impedances next to inverting and non-inverting inputs, and your mismatch is more than ~100-200 Ohms (roughly)

I guess I can't use inverting topology since the preceding stage is the volume controller with the variable output impedance which will vary the gain in inverting topology. What problems do I run into with the impedance imbalance for non-inverting and inverting inputs? Excessive output offset voltage?

Two diode bridges are redundant. There is barely a case where it makes a difference, and this is not the one. Put just one.

The rectifiers can be used in a single bridge configuration using a couple of wires across the appropriate AC inputs if necessary. To make it even easier I can provide solder bridges on the board to help shorting the appropriate traces for this.

Regards,
Oleg
 

Shaq888

Member
2009-02-09 7:23 am
I remember the discussion in the Super Regulator thread which also has remote sensing, where Jan Didden explained that placing large electrolytics at the load defeats the advantages of using the regulators by isolating the load from the regulator. In my understanding regulators "replace" bulk capacitance if they placed close to the point of load. The remote sense (feedback) should take care of lowering the power supply output impedance seen by the load in the audio frequency range or am I wrong here?

Series regulator will pass current trough. You don't want rectified load current to go far, since it increases chances of coupling it into signal chain. It is better to terminate those currents as close as possible. So it is either local bulk caps or parallel regulator.
Sometimes when I can not use a parallel reg but I have to regulate the rail, I put a small resistor in series with the regulator output (order of a fer ohms) and a lager cap next to the IC. This way all high freq current will terminate on the cap. Lower freq do not matter as much.
This barely affect low freq performance since that is the region where opamp has most of PSRR and gain.

I have EMI filter on the preceding stages but I can put it on this board too. I guess you mean small value capacitor to ground between the input resistor and the opamp's non-inverting input to make a low pass RC filter with the corner frequency of around 200~300kHz? Or I need something more sophisticated?

Some RC right next to the input is enough. Don't have to cut too low, unless you have really harsh EMI. In this case a second order filter may be more sensible.


The absence of load capacitance isolation resistor is what was one of the advantages of using the OPA1622 as I thought. If I still have to use it, then it would make it similar to the other more diy friendly headamp designs, e.g. LME49990/LME49600 combo.

Cable is not just a capacitance, it is transmission line which is poorly terminated. This makes things a little bit more complicated. This is not very high speed IC, but with longer cable you can get a first reflection close to opamp crossover frequency.
If you have a spec an with trac gen output you can carefully probe what is happening there when you have cable connected. But don't connect 50Ohm directly. It may provide enough damping and you will not see a thing. Try to probe through something like 1k or higher.
Quite likely a simple RC to the ground may solve issues, if the do exist.


I guess I can't use inverting topology since the preceding stage is the volume controller with the variable output impedance which will vary the gain in inverting topology. What problems do I run into with the impedance imbalance for non-inverting and inverting inputs? Excessive output offset voltage?

In such case you will get an increased distortion level since you have a common mode voltage and unbalanced impedances.
You can you inverting config with a pot, but in this case pot has to be linear type. 10k pot loaded with 2k gives good logarithm approximation.

The rectifiers can be used in a single bridge configuration using a couple of wires across the appropriate AC inputs if necessary. To make it even easier I can provide solder bridges on the board to help shorting the appropriate traces for this.

This is completely up to you, but those extra diodes are most useless thing you can imagine. It is better to remove them and re-layout rectifier as compact as possible to decrease its radiation.
 
Here is one more thing that popped up in the vendor thread to keep in mind about the OPA1622. It has a huge 1uA input bias current! They designed it to be powered by a DAC with balanced impedances on the two inputs. johnc124 said it has no input bias current cancellation. If the chip is run non-inverting and fed directly from a pot there would be significant current through the pot wiper (scratchies) and output offset voltage. If fed from another stage though you should be OK if the chip input impedances are roughly equal.
 
Last edited:

OlegSh

Member
2010-12-01 11:00 am
Germany
Hi agdr,

Thanks for the info! I understand the concern but in my case it does not apply since I use relay based attenuator as the volume control but it might be a problem for someone else. In the worst case AC coupling between the pot and the OPA1622 could help. Introducing another opamp into the design defeats the benefits of OPA1622 for me.

Regards,
Oleg
 

OlegSh

Member
2010-12-01 11:00 am
Germany
Hi Sergey,

Series regulator will pass current trough. You don't want rectified load current to go far, since it increases chances of coupling it into signal chain. It is better to terminate those currents as close as possible. So it is either local bulk caps or parallel regulator.
Sometimes when I can not use a parallel reg but I have to regulate the rail, I put a small resistor in series with the regulator output (order of a fer ohms) and a lager cap next to the IC. This way all high freq current will terminate on the cap. Lower freq do not matter as much.
This barely affect low freq performance since that is the region where opamp has most of PSRR and gain.

This confuses me even more. It sounds like the regs are completely useless:)

Cable is not just a capacitance, it is transmission line which is poorly terminated. This makes things a little bit more complicated. This is not very high speed IC, but with longer cable you can get a first reflection close to opamp crossover frequency...

Quite likely a simple RC to the ground may solve issues, if the do exist.

Do you suggest to add a Zobel network at the output? So far I haven't seen it in the other headphone amplifier designs.

Regards,
Oleg
 

Shaq888

Member
2009-02-09 7:23 am
This confuses me even more. It sounds like the regs are completely useless:)

That is why I said earlier that I would rather put just an extra filter or nothing at all :)

See these spectra? I got them with 1 ohm resistors in lines with the bench PSU. And I have got a few dB improvement on even harmonics when I managed to get electrolytics a few mm closer. Perhaps an improvement was more due to reduced coupling between caps terminals and a signal path, but still an improvement.

Localizing currents is a big thing when you what to achieve a low distortion level, especially at a higher freq end.

Combination of series and parallel regs can be a very good solution, but I think it is redundant in this case.

Do you suggest to add a Zobel network at the output? So far I haven't seen it in the other headphone amplifier designs.

This is the reason why you occasionally hear people say that cables make a huge difference. Shorter/longer cable/ +/-100pF and you've got a generator :)
In reality not a lot of people have means to do this kind of measurement, so no one really pays an attention to this problem.

Zobel might be OK, something like 51Ohm 4.7nF (NP0/C0G). Also you can try modify a loop gain a bit, moving a crossover freq lower. But first of all you need to be able to see if there is a problem and how it changes with modifications.
 
Last edited:
Hi agdr,

I already soldered the regulators on four dual polarity boards with success. I fried one board at first and thrown away two chips while learning but in the end it worked with the solder paste and the hot air solder station for 40$ from ebay. The OPA1622 is a bit more challenging but I intentionally put a big via in the thermal pad to heat it from below. I guess it should work this way.

Regards,
Oleg

Regards,
Oleg
 

Shaq888

Member
2009-02-09 7:23 am
I have to say that common mode distortion, even with imbalanced impedances, are not bad at all. However, you sticc can see the difference. There are a few spectra. VS=+/-10V, G=+1, 3k3 load, 2Vrms(+6dBVrms) signal, 1kHz, 10kHz and 20kHz. In both cases source resistance is 1k. In one of them I balanced resistance next to inverting input, in other I connected it directly to the output.
http://i.imgur.com/Ykcs0Pk.png
http://i.imgur.com/6Or82PF.png
http://i.imgur.com/LjGGEd8.png
http://i.imgur.com/JHC6MUY.png
http://i.imgur.com/bSf7yOT.png
http://i.imgur.com/r1qnaPO.png
 
Last edited:
Thanks for posting your measurements, Sergey! Now I can sleep well:)

Agdr, I would not rely entirely on the surface tension to position the parts during soldering. It will happen for the chip especially with the thermal pad but bulky two terminal parts may still be misplaced after reflow. So it is better to put them straight initially.
 
Last edited:

Shaq888

Member
2009-02-09 7:23 am
No problems.

If you are going to use it with pot in non-inv config, you can do some optimizations.
Use low value pot, like 4.7k.
Than took the highest possible output resistance, when it is in the middle, and divide it by two. Than add resistance of EMI filter. This is approx resistance you need to have connected to the inv. pin to have OK-ish imbalance in any position of the pot. Also you can take a bit higher value, if you think that min and max positions of the pot wont be used a lot. Also in min position distortions will not be high due to low common mode signal.