• Disclaimer: This Vendor's Forum is a paid-for commercial area. Unlike the rest of diyAudio, the Vendor has complete control of what may or may not be posted in this forum. If you wish to discuss technical matters outside the bounds of what is permitted by the Vendor, please use the non-commercial areas of diyAudio to do so.

Retro - A fully symetrical phono stage with RIAA filter

Hi Folks,

I can't count how many times I have been asked when I would design a phono stage. Well I really don't know why it took me so long. But here it is.

The Retro is a unique design which uses an instrumentation amplifier front end with a 47K input impedance so as not to overload MM magnetic cartridges. It is equally well suited to MC cartridges because its gain can be easily increased to over 60db with little penalty.

The RIAA filter is reasonably accurate, at least in simulation. It should also be in practice since 1% caps and .1% resistors are used.

The fully symmetrical nature of the circuit means that distortion is low and common mode input noise is rejected very well. You can still use normal RCA inputs with one end of the cartridge output grounded. I have been using it with the cartridge floating. I use XLR with pin 1 connected to the retro GND and pins 2 and 3 connected only to the cartridge outputs. One per channel.

I used a "H" network to raise the gain of the fully differential stage without using too high value feedback resistors or overly loading the instrumentation amp. This lowers the noise and distortion a bit.

The output is balanced, but you are free to only take one side if you like and make it single ended.

Joshua Tree works wonderfully at the output to make a complete preamp.

It sounds awesome. :)

Here is the filter characteristic:
Code:
  step	v(out+,out-) / ref	at
     1	(19.3351dB,28.4648°)	20
     2	(19.0946dB,25.144°)	23.7841
     3	(18.7762dB,21.4673°)	28.2843
     4	(18.3631dB,17.4889°)	33.6359
     5	(17.8403dB,13.3026°)	40
     6	(17.1968dB,9.04005°)	47.5683
     7	(16.428dB,4.85961°)	56.5685
     8	(15.5377dB,0.927718°)	67.2717
     9	(14.5371dB,-2.60171°)	80
    10	(13.4436dB,-5.60337°)	95.1366
    11	(12.2785dB,-7.98836°)	113.137
    12	(11.0649dB,-9.70432°)	134.543
    13	(9.82634dB,-10.7314°)	160
    14	(8.58624dB,-11.0782°)	190.273
    15	(7.3677dB,-10.7805°)	226.274
    16	(6.19321dB,-9.90373°)	269.087
    17	(5.08393dB,-8.54749°)	320
    18	(4.05806dB,-6.85014°)	380.546
    19	(3.12849dB,-4.98767°)	452.548
    20	(2.30022dB,-3.16429°)	538.174
    21	(1.5685dB,-1.59371°)	640
    22	(0.918372dB,-0.474936°)	761.093
    23	(0.325981dB,0.0312881°)	905.097
    24	(-0.238887dB,-0.179877°)	1076.35
    25	(-0.809554dB,-1.14756°)	1280
    26	(-1.419dB,-2.84291°)	1522.19
    27	(-2.0965dB,-5.17411°)	1810.19
    28	(-2.86451dB,-7.99901°)	2152.69
    29	(-3.73654dB,-11.1454°)	2560
    30	(-4.71644dB,-14.4359°)	3044.37
    31	(-5.79946dB,-17.7117°)	3620.39
    32	(-6.9747dB,-20.8485°)	4305.39
    33	(-8.22801dB,-23.7636°)	5120
    34	(-9.54455dB,-26.4125°)	6088.74
    35	(-10.9105dB,-28.7821°)	7240.77
    36	(-12.3142dB,-30.881°)	8610.78
    37	(-13.746dB,-32.7313°)	10240
    38	(-15.1984dB,-34.3627°)	12177.5
    39	(-16.6659dB,-35.8081°)	14481.5
    40	(-18.1443dB,-37.101°)	17221.6
    41	(-19.4245dB,-38.1167°)	20000

Please let me know if you all are still interested.

Enjoy the attachments.

Cheers!
Russ
 

Attachments

  • schematic.pdf
    27.7 KB · Views: 4,440
  • pcb.jpg
    pcb.jpg
    192.9 KB · Views: 8,152
  • RIAA.pdf
    26 KB · Views: 2,376
Last edited:
Russ,

Let me make sure I understand how this works; please correct any mistakes.

IC2.1 and IC2.2 form an instrumentation amplifier with flat frequency response. As always, common mode gain is 1, while normal mode gain is given by (R1+R3)/R2 - I think. What device is used for IC2? I am surprised you can get a good noise match for impedances ranging from 47K (MM at 20kHz) to a few ohms (MC), unless you use a FET input device, but few of these have low en.

Equalisation is implemented by a passive network R5, R6, R4, C1, C2, loaded by R9 and R10. The gain of the instrumentation amp needs to exceed the attenuation of the network at 20kHz, otherwise the front end doesn't capture the noise figure.

The final stage built round IC1, removes common mode, and adds flat frequency response gain to make up the losses in the passive stage.

My own preference is to use less gain, by splitting the equalisation, and putting the 75us pole in the position where your network is, and moving the low and mid frequency time constants (3180us/318us) into the second stage feedback networks. In a differential design, this puts capacitor matching into the critical region that determines CMRR, so I see why you went with the topology chosen. The concern with all passive eq setups is that you are trapped in a narrow window between too little first stage gain (giving HF noise compromise) and too much (giving poor HF headroom).

I would be interested to see the component values, so I can think about the compromise you have picked.
 
Preliminary cct values

Hi PD,

Great questions from you as I would expect. :)

You have a good grasp of the circuit. Nothing I can find to correct you on.

I will post the values I am currently using in my test rig below. Keep in mind. This cct is still in development. So now is the perfect time to get good input from folks like you. I don't have PCBs made yet, so plenty of time for change.

Right now I am using another PCB that I have modified to fit the bill (MESH) to do my prototype testing.

For now here are some numbers.

IC1 is OPA1632, IC2 is OPA2228(Though I am open to suggestions). AD8620 is one I will also try. There is also a new dual JFET input opamp from TI (I forget the part at the moment) that looks promising.

All resistors are .1% where good matching is required.

R1/R3 22K R2 is 1K. This yields a differential gain of 33db for the instrumentation amp.

This RIAA filter/diff input block is as follows:
R5/R6 are 493R
R9/R10 are 1.5K
R4 is 105R
C1 is 1uf 1%
C2 is 3uf 1%

The feedback H network is as follows:
R11-R15 are all 10K for MM. (system gain is 39.1 db at 1khz)
R13 is 470 for for low output MC. (system gain is 59.9db at 1khz)
R13 is 2.21K for high output MC. (system gain is 47.9db at 1khz)

As I said before. These are all preliminary values. But they seem to be working very well for me so far.

Cheers!
Russ
 
Last edited:
I'm not sure what R2 and R13/R20 are doing.
R2 is the gain or shunt resistor for the instrumentation amp. As PD correctly stated the gain of that block is (R1 + R3) / R2.

R13 is the shunt for the "H" network.

An "H" network is to a fully symmetrical amp what a "T" network is an ordinary inverting amplifier.

The "H" network basically allows me high gain while maintaining reasonable impedance into the filter.

Cheers!
Russ
 
Last edited:
Right one problem straight away - for low output MC, you need really low impedance in the feedback net round IC2, otherwise the Johnson noise swamps everything.

Remember, for a good MC stage, we want to target well under 1nV/rtHz at the input. This is equivalent to the noise in a 50 Ohm resistor. Your current values will be way too noisy.

So I think for MC you need to use much lower values for R1-3; in order to not run out of current swing this probably means running with a higher gain at this stage. I would suggest say R1 and R3 2K2, R2 10 Ohms, giving an extra 20dB odd of gain; doing it here means you can leave the 2nd stage alone.

More 1st stage gain is also needed to avoid the 2nd stage dominating the noise floor at HF; note that the RIAA curve drops nearly 40db between 20Hz and 20kHz. This means an all passive design needs to have at least 40dB of front end gain, to avoid compromising the noise figure, so you probably need to do this for MM anyhow.

The scary bit is what happens on big HF transients - something like a Shure V15 can track over 70cm/s in the upper midrange, giving well over 50mV. With x200 gain, this is well over 1Vrms before the eq net, so we don't have enormous factors of headroom, but OK.

Wacky choices (like say a Koetsu black into a x20 transformer) can give much larger levels still, but probably don't track such high velocities.

I don't know the 2228 - is it happy driving several mA into the feedback net?
 
One testing issue I have is I don't own any low output moving coil cartridges. :)

I have a few MM types and a Denon DL160 high output MC.

All of those have worked very well.

DL103 is cheap and widely liked. I have an AT OC9 which is mid priced, and I think very good, but some people think too thin.

For testing purposes one of the low output (and moderate cost) Ortofons is ideal like an MC20; if a stage is quiet enough to make that usable, it is generally useful. My OC9 has a slightly higher output (600uV) which makes it easier to hide noise levels.
 
HI PD all good points.

And here is something I had thought about.

I am thinking I may make two versions.

A MC version with a op1632 front end. This cct would have basically close to 0R input impedance. It would really be an I/V stage. This of course would not be at all optimal for MM cartridges. But great for MC.

I can certainly go higher gain with the front end with the cct shown and with lower Rs, but I just have not tried that yet.

It is a trade off between being able to support both MM and MC on one board, or best performance for each MM and MC using two boards.

I think my inclination right now is the latter.
 
Last edited:
One possible compromise that I have used (about 25 years ago!) is two frontends , and switch between them.

In my case, there were flat frequency response gain boards - from memory something like 20dB for MM, and 46dB for MC. The output of a switch went to the eq board; this had a passive 75us low pass, and the other time constants in the feedback loop of the second gain stage.

Another point - the design as shown is DC coupled. Depending on arm+cartridge and vinyl flatness, you can get some scary big subsonics. My own feeling is that you want at least one high pass time constant somehere; I went for about 10mS (16Hz -3dB) as it loses little audible, but is rolling off fairly well by the time you get to warp related stuff (harmonics of 0.5Hz). Some designs have multiple time constants; this trades phase shift for better supression of subsonic manure.
 
PD,

Another excellent point on the high pass.

One way I had though about tackling this was to apply a symmetrical integrator (servo) with a relatively high corner. I was thinking 10-16hz.

This would have two advantages. It should remove any differential DC introduced at either the first or second stages due to input currents, part matching ,etc... and it would act as a high pass.

Any thoughts on this approach?
 
Should work, and avoids the need for large matched caps to protect CMRR at 50/60Hz.

As the gain at DC for an MC is very high (60dB), even normal input offsets of a few 100uV or so turn into worryingly large values unless you do something like it.

Never done it that way myself. I guess there are some issues to check in terms of startup behaviour and recovery from overload, but a short time constant should fix pretty much all ills.
 
Retro to USB

I have a large vinyl record collection dating back to my first LP purchase in 1964. So I'd like a way to take the Retro output, digitize it (possibly through the ESS 9102 ADC), and then input it into my PC through the USB port. The combination of Retro and a good ADC, would most likely make a large improvement on the digitized recordings I've made with a cheap $99 USB digitizer that I currently have. Russ, any plans to go in this direction, or am I just being nuts?

RossG
 
I have a large vinyl record collection dating back to my first LP purchase in 1964. So I'd like a way to take the Retro output, digitize it (possibly through the ESS 9102 ADC), and then input it into my PC through the USB port. The combination of Retro and a good ADC, would most likely make a large improvement on the digitized recordings I've made with a cheap $99 USB digitizer that I currently have. Russ, any plans to go in this direction, or am I just being nuts?

RossG


The ES9102 is on the agenda. It will be its own module. It will would work just fine with this cct.