ken kantor
Fowl Humor
+1 on this.
I will take a look through some dusty lab notes of mine to find the inverse-RIAA I used to rely on for development.
-k
Sonic questions about self built RIAA print
- Thread starter Spooter
- Start date
+1 on this.
I will take a look through some dusty lab notes of mine to find the inverse-RIAA I used to rely on for development.
-k
You wrote about "dealers and customers were complaining about distortion problems. It didn't take long for Stan and I to discover that the amp driving the passive network was being over-driven despite careful attention to gain staging"
Is this still the problem with the 640p which came on the market or have you made modifications since?
ken kantor
Fowl Humor
I am embarrassed to say that I was confused by another thread I was reading here.
The amp I was relating the story about was the Cambridge Audio P35, from back in the mid-1980's. The mods we made were immediately put into the production units, and solved the marketplace issue.
Sorry for the confusion!
-k
Attachments
Last edited:
@mr kantor
The cardridge loading is not the cause of the issue since the actual input resistor R1 is 47k, not 49.9 and capacitor C1 is not placed. Rega does not specify anything but tracking force, so I have to assume this is the proper setup for the rega bias MM cardridge. (these values were posted by 'Spooter' 17/11) I am unaware of a different loading requirement for this cardridge.
Cable capacitance is probably 100-200pF (needs to be measured) If required a capacitor can be placed, but a rega specification is required then.
I think Lipshitz is pretty clear indeed about the existence of the HF zero for the non-inverting topology:
"There is an additional unavoidable high-frequency turnover with time constant T6 (corresponding to a frequency (f6, say) which appears in Fig. 3 even when R3 = 0. This departure from the ideal RIAA deemphasis curve does not arise in the inverting case (Fig. 2), unless we deliberately set R3 ≠ 0."
- Which "change to the spec" is going to change this?
The only way I can think of which minimizes (not prevents) the effect is to use a single stage with the RIAA section in the feedback and at least 60dB DC gain. This zero (for a unity gain stable opamp) is then at 200kHz and probably of little importance to audio. On the other hand that will:
- 'make the op amp work hard (talking 'loop gain' now. At 20kHz, the closed loop gain is then 10x, while in 'my' circuit 1st stage it is 25x. Not a really big difference)
- either the high frequency load impedance will be even lower ( the resistor on the negative input is reduced, bad for the opamp)
- or the impedance of the RIAA section needs to be increased (increased noise)
For example: for 60dB DC gain, you need 1:1000 resistors. 100 ohm on the inverting input to ground then requires 100k in the feedback (and matching capacitors). 100 Ohm HF load impedance will not make the op amp happy. Increasing to 1k/1M is bad for the noise specs.
Another thing I worry about is capacitive loading of the op amp. I know many power amplifiers have problems with reactive/complex loads. I wonder why op amps would be more tolerant...
In short: I'm not sure putting the RIAA section in the loop is simply 'better'.
@ConradH
Spice seems to be the favourite tool for many, but personally I prefer matlab over spice for simulation. It is what I used to make the transfer plots
I think it is better to simulate in a package you know than needing to (re-)learn something else which does pretty much the same thing. The difference is that instead of working with components I have to formulate complex impedances and frequency transfer functions, so the approach is a bit more abstract/one more level removed from the physical level, but i am confident Matlab (and I) can do the job.
The major advantage for me is that I have to calculate the analytic expressions for the poles and zeros and thus get a better insight in how the R and C values effect each. I prefer this over changing one of the values and then simulate to see the effect, which I would call the 'magical' approach.
It is one of the things I do for a living, so i consider myself experienced enough to simulate a 4 component RC network. This is confirmed by the fact that matlab returns a flat response when simulating a theoretical inverse RIAA (defined in time constants, not a circuit) in series with the transfer function of Jung's RIAA circuit+component values. If this is correct (=flat), surely I can simulate the effect of small deviations from Jung's component values.
I understand an inverse RIAA network and a square wave generator give a fast test result by just looking at the edges. I am perfectly happy though measuring the preamp on its own with a sine input. There are tables on the web and in the op amp book by Jung (and my simulation of the circuit which also gives the phase resonse) which give the desired gain at any frequency.
Apart from being more work and thus an increased possibility for errors, what serious 'pitfalls' do you expect?(a single error in a series of 10 is easy to spot and then you repeat that measurement. On the other hand: if you do a single measurement which is wrong....you may never notice)
As far as the hagtech inverse riaa is concerned: I think the design procedure as described in http://www.hagtech.com/pdf/riaa.pdf is a bit questionable: in the spice netlist the circuit is deliberately split in two sections which do not interact. The point is: in the circuit the interaction does happen, why would you not put that in your model if you're running it through a computer anyway?
The load impedance it presents to the source seems to be lowish (and frequency dependent) considering consumer CD players which as far as I know typically have an output impedance of up to 500 Ohm. (This is relevant when it is used for breaking in equipment as advertised)
The 3.18µs pole is debatable (Douglas Self disputes it, although it probably doesn't matter if you use a 400Hz square wave test signal. Don't be surprised though if 20kHz sine transfer is 0.8dB off, see the green line)
Finally I think it is 'strange' 2 capacitors of 3.3nF are used to make 6.2nF(C2!) Why not use something like 4.7nF + 1.5nF. Fortunately the resulting mid-boost is only 0.4dB.
The bottom line is: I will measure to check the RIAA conformance when we meet again, but so far my design philosophy is not seriously challenged.
The cardridge loading is not the cause of the issue since the actual input resistor R1 is 47k, not 49.9 and capacitor C1 is not placed. Rega does not specify anything but tracking force, so I have to assume this is the proper setup for the rega bias MM cardridge. (these values were posted by 'Spooter' 17/11) I am unaware of a different loading requirement for this cardridge.
Cable capacitance is probably 100-200pF (needs to be measured) If required a capacitor can be placed, but a rega specification is required then.
I think Lipshitz is pretty clear indeed about the existence of the HF zero for the non-inverting topology:
"There is an additional unavoidable high-frequency turnover with time constant T6 (corresponding to a frequency (f6, say) which appears in Fig. 3 even when R3 = 0. This departure from the ideal RIAA deemphasis curve does not arise in the inverting case (Fig. 2), unless we deliberately set R3 ≠ 0."
- Which "change to the spec" is going to change this?
The only way I can think of which minimizes (not prevents) the effect is to use a single stage with the RIAA section in the feedback and at least 60dB DC gain. This zero (for a unity gain stable opamp) is then at 200kHz and probably of little importance to audio. On the other hand that will:
- 'make the op amp work hard (talking 'loop gain' now. At 20kHz, the closed loop gain is then 10x, while in 'my' circuit 1st stage it is 25x. Not a really big difference)
- either the high frequency load impedance will be even lower ( the resistor on the negative input is reduced, bad for the opamp)
- or the impedance of the RIAA section needs to be increased (increased noise)
For example: for 60dB DC gain, you need 1:1000 resistors. 100 ohm on the inverting input to ground then requires 100k in the feedback (and matching capacitors). 100 Ohm HF load impedance will not make the op amp happy. Increasing to 1k/1M is bad for the noise specs.
Another thing I worry about is capacitive loading of the op amp. I know many power amplifiers have problems with reactive/complex loads. I wonder why op amps would be more tolerant...
In short: I'm not sure putting the RIAA section in the loop is simply 'better'.
@ConradH
Spice seems to be the favourite tool for many, but personally I prefer matlab over spice for simulation. It is what I used to make the transfer plots
I think it is better to simulate in a package you know than needing to (re-)learn something else which does pretty much the same thing. The difference is that instead of working with components I have to formulate complex impedances and frequency transfer functions, so the approach is a bit more abstract/one more level removed from the physical level, but i am confident Matlab (and I) can do the job.
The major advantage for me is that I have to calculate the analytic expressions for the poles and zeros and thus get a better insight in how the R and C values effect each. I prefer this over changing one of the values and then simulate to see the effect, which I would call the 'magical' approach.
It is one of the things I do for a living, so i consider myself experienced enough to simulate a 4 component RC network. This is confirmed by the fact that matlab returns a flat response when simulating a theoretical inverse RIAA (defined in time constants, not a circuit) in series with the transfer function of Jung's RIAA circuit+component values. If this is correct (=flat), surely I can simulate the effect of small deviations from Jung's component values.
I understand an inverse RIAA network and a square wave generator give a fast test result by just looking at the edges. I am perfectly happy though measuring the preamp on its own with a sine input. There are tables on the web and in the op amp book by Jung (and my simulation of the circuit which also gives the phase resonse) which give the desired gain at any frequency.
Apart from being more work and thus an increased possibility for errors, what serious 'pitfalls' do you expect?(a single error in a series of 10 is easy to spot and then you repeat that measurement. On the other hand: if you do a single measurement which is wrong....you may never notice)
As far as the hagtech inverse riaa is concerned: I think the design procedure as described in http://www.hagtech.com/pdf/riaa.pdf is a bit questionable: in the spice netlist the circuit is deliberately split in two sections which do not interact. The point is: in the circuit the interaction does happen, why would you not put that in your model if you're running it through a computer anyway?
The load impedance it presents to the source seems to be lowish (and frequency dependent) considering consumer CD players which as far as I know typically have an output impedance of up to 500 Ohm. (This is relevant when it is used for breaking in equipment as advertised)
The 3.18µs pole is debatable (Douglas Self disputes it, although it probably doesn't matter if you use a 400Hz square wave test signal. Don't be surprised though if 20kHz sine transfer is 0.8dB off, see the green line)
Finally I think it is 'strange' 2 capacitors of 3.3nF are used to make 6.2nF(C2!) Why not use something like 4.7nF + 1.5nF. Fortunately the resulting mid-boost is only 0.4dB.
The bottom line is: I will measure to check the RIAA conformance when we meet again, but so far my design philosophy is not seriously challenged.
Attachments
As found on the internet:
Type: Moving Magnet
Output Voltage: 6.8mV
Tracking Force: 1.75g
Load Impedance: 47k ohms
Stylus Type: Elliptical
http://www.soundssupreme.co.uk/rega_bias.html
Type: Moving Magnet
Output Voltage: 6.8mV
Tracking Force: 1.75g
Load Impedance: 47k ohms
Stylus Type: Elliptical
http://www.soundssupreme.co.uk/rega_bias.html
ken kantor
Fowl Humor
- Got it! The input loading is close enough for now.
- What I was trying to say about the HF zero, and what Stanley said, is that it is not an issue at all, provided your loop gain is OK.
- It is not clear to me how putting the network in the feedback loop, could possibly increase the HF loop gain. If you reduce the CL gain, you increase the L gain, at any given frequency. (BTW- I would say that many of the best RIAA preamps put the network in the feedback, so it must be possible to do correctly, no?)
- It really isn't productive to discuss the sonics of the preamp before measurements are done. Something could be built wrong, and open listening tests are notoriously unreliable, even with panels and protocols. Once we see both the preamp and the cartridge/preamp responses, (beware cheap test discs!), then we can discuss sonics meaningfully.
- When doing the measurements, keep a close eye on the DC servo. Is it leaking audio? Modulating the bass? Pumping?
- Why the worries about impedance? Decent opamps are just fine running somewhat below 1KHz. Am I missing something?
- Of course, you know what you are doing, but I still recommend building yourself an inverse network. It's easy, and I doubt anyone who has done this more than a few times would want to work without it. For example, it's a total pain to do IMD or complex overload tests on a bare input.
To be sure, I think your design is "sound." My comments are only being made to help track down possible performance-related sonic problems, and to pass along received wisdom.
-k
Last edited:
ken kantor
Fowl Humor
Yeah, this is a pretty minimalist design, as is. With the exception of the DC servo, I see nothing extraneous.
-k
ken kantor
Fowl Humor
Hey, I had a chance to read your reply again, more carefully. A few brief comments:
- I see you were talking about CL gain, not loop gain. So, we are in general agreement. I still think that it is "better" to have more loop gain than less, but the practical benefits might be nil. What's the OL gain of the opamp you are using up around 20KHz?
- The problem with opamps and capacitive loads is that the load can force the output phase up near the unity gain frequency to rotate 90 degrees. Having the caps in the feedback loop is a whole different animal. In this case, the caps are not to ground, and so can't easily force the LZ output. More important, the feedback caps mean that the CL gain is far, far, below unity up near the OL unity gain bandwidth, so oscillation is not possible. Such an amp is generally highly stable, much more stable than a garden variety non-inverting amp. Make sense?
-k
Last edited:
Reaction from the designer and builder:
Spooter (topicstarter) tested the setup with a 3rd party RIAA preamplifier which turned out to have the same 'problem'. So 'technically' the case can be closed.
Call me lazy for not building an inverse RIAA, but IMO this was a good move because testing with a 3rd party reference is always better than constructing your own device with the same possibility for logic errors as may have existed in the RIAA preamplifier in the first place.
My worries regarding impedance come from some (superficial) reading of some literature by Douglas Self (small signal audio design). I visited an AES masterclass in Amsterdam where he talked about power amplifiers. As far as I can judge (I am an electrical engineer, though not specifically in the audio field) he knows what he is talking about. He is also a proponent of placing the RIAA network inside the loop, you might like that.
In this book there are a lot of design ideas to minimize noise, increase voltage headroom etc. One of the measures he takes in the RIAA design is to decrease the feedback impedance (for "reduced Johnson noise"): the resistance from the negative input to ground goes from 1k to 220.
In the 'op amp applications' handbook Jung also gives some designs with RIAA feedback where the negative input to ground resistance is typically 100-200 ohm.
These are not values 'somewhat' below 1k, these are significantly, up to 1 order of magnitude below 1k.
The op amps I used in the RIAA preamp is a OP37 (3nV/sqrt(Hz) @ 1kHz noise, slew rate 17V/µs, 126dB CMRR, 10µV offset, 63MHz gain bandwidth).
The datasheet of this amplifier has a nice plot of 'maximum output voltage vs load resistance' and it clearly shows that below 1k the output swing decreases seriously.
The loading suggested by Self and Jung (<200) is thus too high for this op amp, and this is a very good op amp!
A typical opamp any audio designer would use is the 5532 (5nV/sqrt(Hz) noise, slew rate 9V/µs, 100dB CMRR, 0.5mV offset, 10 MHz gain bandwidth). For all its merits, the 5532 opamp is from 1979 and seems to be outperformed on all these specs by the newer op37. The 5532 datasheets (TI or National) do not provide data on output loading. Based on the previously observed differences on critical specs such as gain-bandwidth product and slew rate, I find it hard to believe that the 5532 is happy to drive sub 1k loads, especially at high frequencies (>10kHz?).
Am I worrying too much?
Did some more research on the OP37 and 5532. Have the following corrections to myself:
- The OP37 is as ancient as the 5532. (Is there really nothing newer?)
- I found out Douglas Self took the time to do some proper measuring. According to these measurements the 5532 is in practice less noisy and has less THD as the OP27, which is closely related to the OP37.
He does however never load the thing below 500 ohm during these measurements, which he calls it's 'maximum'. No idea where he got that information.
Did I say 'national'? I meant On semi. They also provide an application note AND8177 which has only 49.9 ohm from the negative input to ground, thus a high freq. load of only 50 ohm.
Am I the only one? I work mostly in power electronics. You don't get to neglect current rating in power electronics.
Spooter (topicstarter) tested the setup with a 3rd party RIAA preamplifier which turned out to have the same 'problem'. So 'technically' the case can be closed.
Call me lazy for not building an inverse RIAA, but IMO this was a good move because testing with a 3rd party reference is always better than constructing your own device with the same possibility for logic errors as may have existed in the RIAA preamplifier in the first place.
My worries regarding impedance come from some (superficial) reading of some literature by Douglas Self (small signal audio design). I visited an AES masterclass in Amsterdam where he talked about power amplifiers. As far as I can judge (I am an electrical engineer, though not specifically in the audio field) he knows what he is talking about. He is also a proponent of placing the RIAA network inside the loop, you might like that.
In this book there are a lot of design ideas to minimize noise, increase voltage headroom etc. One of the measures he takes in the RIAA design is to decrease the feedback impedance (for "reduced Johnson noise"): the resistance from the negative input to ground goes from 1k to 220.
In the 'op amp applications' handbook Jung also gives some designs with RIAA feedback where the negative input to ground resistance is typically 100-200 ohm.
These are not values 'somewhat' below 1k, these are significantly, up to 1 order of magnitude below 1k.
The op amps I used in the RIAA preamp is a OP37 (3nV/sqrt(Hz) @ 1kHz noise, slew rate 17V/µs, 126dB CMRR, 10µV offset, 63MHz gain bandwidth).
The datasheet of this amplifier has a nice plot of 'maximum output voltage vs load resistance' and it clearly shows that below 1k the output swing decreases seriously.
The loading suggested by Self and Jung (<200) is thus too high for this op amp, and this is a very good op amp!
A typical opamp any audio designer would use is the 5532 (5nV/sqrt(Hz) noise, slew rate 9V/µs, 100dB CMRR, 0.5mV offset, 10 MHz gain bandwidth). For all its merits, the 5532 opamp is from 1979 and seems to be outperformed on all these specs by the newer op37. The 5532 datasheets (TI or National) do not provide data on output loading. Based on the previously observed differences on critical specs such as gain-bandwidth product and slew rate, I find it hard to believe that the 5532 is happy to drive sub 1k loads, especially at high frequencies (>10kHz?).
Am I worrying too much?
Did some more research on the OP37 and 5532. Have the following corrections to myself:
- The OP37 is as ancient as the 5532. (Is there really nothing newer?)
- I found out Douglas Self took the time to do some proper measuring. According to these measurements the 5532 is in practice less noisy and has less THD as the OP27, which is closely related to the OP37.
He does however never load the thing below 500 ohm during these measurements, which he calls it's 'maximum'. No idea where he got that information.
Did I say 'national'? I meant On semi. They also provide an application note AND8177 which has only 49.9 ohm from the negative input to ground, thus a high freq. load of only 50 ohm.
Am I the only one? I work mostly in power electronics. You don't get to neglect current rating in power electronics.
Last edited:
ken kantor
Fowl Humor
I'm glad the problem is solved!
Honestly, and no disrespect to anyone here or in my past.... I learned a few years in to save the listening tests until the last phase of development. The headaches I have had because my boss wanted to hear the new prototype and, "understood it wasn't finished"........!
I understand your point about impedance now, but Self and Jung are pushing impedance down farther than it needs to go for an op amp circuit, IMO. Plus, the EQ curve mitigates the problem in practice, due to the reduced swing requirements at increasing frequencies. Still, you are correct here. For a few years, I worked on a series of preamps that stuck a low noise FET in front of op amp to try and optimize the impedances for both feedback and input noise figure. At this point, I think I would just raise the feedback Z rather than go to that trouble, given improvements in output stage drive capabilities. But, that is purely a judgment call.
Anyway, hope you enjoy the pre. Let me know how it turns out, if you think of it.
-k