Very low noise op amp recommendations?

[bikerhifinut]

It seems the decision is made for me and as I have a couple of 5534 I don't need to spend a quid!
Am I right in thinking that a TL07x is better suited for the second stage which will have a gain of up to 800x at certain frequencies or does the better current drive of the 5532 win here?
The gain issues in stage 1 aren't a problem as its got 20dB gain there. So at a pinch a 5532 works too.
Its not cutting edge but I wanted to get the best practical result from the circuit.

by the way I assume its the same m hennessy whose website I have perused when researching the gainclone amps?

A.

[mhennessy]

I'm not sure why you're worried about the 5532 at high gains - it should be fine. I did see your earlier comment about instability at high gains, but honestly, I would suspect the problem was something other than the op-amp.

I'd never pick a TL07x over an NE553x unless there was a really good reason, such as the DC input currents of the 553x being too high for the application circuit.

BTW, x800 is nearly 60dB - are you sure about that? I haven't seen your circuit so far, I don't think...

Mark

[bikerhifinut]

That floored me initially, but its apparently the maximum gain at one end of the RIAA curve, I had to ask about it but its where the LF correction is at maximum and then the capacitor on the LF filter starts to conduct which takes the AC gain down. Bad explanation but I kind of understand the gist.
Yeah a flat gain of 60dB would mean 48V from 5mV after the initial 20dB gain and that was making my head spin for a while.

Andy

[Radio Wrangler]

The external compensation capacitor is just a bit more precise than internal ones can be, so the phase and gain responses are a bit tighter controlled. It makes a difference when you're pushing these things to the edge.... which an RIAA circuit isn't. Handy when you're doing state-variable analogue filters.

David

[Radio Wrangler]

The compensation capacitor takes the forwards gain down above a turnover frequency.

The RIAA curve requires falling closed-loop gain with frequency, so the two things track together, which is nice.

The compensation pole breakover frequency is between 100 and 300Hz depending on choice of compensation capacitor, so the amplifier gain/phase are reasonably flat at the bottom of the RIAA curve frequency range.

As opamp applications go, this is a nice friendly one. There should be no trouble with instability if good practice is followed.

David

[mhennessy]

Originally Posted by bikerhifinut

That floored me initially, but its apparently the maximum gain at one end of the RIAA curve, I had to ask about it but its where the LF correction is at maximum and then the capacitor on the LF filter starts to conduct which takes the AC gain down. Bad explanation but I kind of understand the gist.
Yeah a flat gain of 60dB would mean 48V from 5mV after the initial 20dB gain and that was making my head spin for a while.

Yes, the bass end is about 20dB up, relative to 1kHz. So if your mid-band gain is 40dB, then that gets you there. But I asked because you mentioned that you've got two gain stages. So I presume your first stage has little/no gain at LF then?

If you have 40dB gain at 1kHz, then 5mV in gives you 500mV out. The Self pre-amp has a little less gain than that - around 50dB IIRC, but don't quote me on that - but all the gain comes from a single 5534. As I mentioned earlier, it could do with an extra 10dB* of gain, but that's easily arranged with a simple x3 amplifier after the RIAA stage. I've never bothered because there's enough "gain in hand" in the rest of the system - you just have to turn the volume up a bit for records.

* My off-the-cuff value of 10dB comes about because my preamp has a volume control calibrated in dB, and that's roughly how much louder CDs are

Take heart in knowing that the 553x op-amps are often used in mic preamps that have 60dB or more of gain. With appropriate attention to detail - decoupling, layout, etc - they are more than capable of doing that

Mark

[bikerhifinut]

And as I watch "You only live twice" I chuckle at davids tagline......

[bikerhifinut]

Originally Posted by mhennessy

Yes, the bass end is about 20dB up, relative to 1kHz. So if your mid-band gain is 40dB, then that gets you there. But I asked because you mentioned that you've got two gain stages. So I presume your first stage has little/no gain at LF then?

If you have 40dB gain at 1kHz, then 5mV in gives you 500mV out. The Self pre-amp has a little less gain than that - around 50dB IIRC, but don't quote me on that - but all the gain comes from a single 5534. As I mentioned earlier, it could do with an extra 10dB* of gain, but that's easily arranged with a simple x3 amplifier after the RIAA stage. I've never bothered because there's enough "gain in hand" in the rest of the system - you just have to turn the volume up a bit for records.

* My off-the-cuff value of 10dB comes about because my preamp has a volume control calibrated in dB, and that's roughly how much louder CDs are

Take heart in knowing that the 553x op-amps are often used in mic preamps that have 60dB or more of gain. With appropriate attention to detail - decoupling, layout, etc - they are more than capable of doing that

heres the circuit.

[bikerhifinut]

So is the EQ tweakable? I figure the values are as high as they are because the bottom resistor in the feedback network is only 1k. But at least they are easily sourced preferred values. Theres also a bit of decoupling needs doing on stage 1 between positive and the top of R1, something like a 4.7k and a 47uF maybe?
And is a resistor on the output a good idea for HF stability? Most commercial preamps seem to put something like a 500R to 1k here.

A.

[Synchrodyne]

Regarding the quest for very low noise, the work done by H.P. Walker (HPW) nearly 50 years ago (Wireless World (WW) 1972 May p.233ff showed that with a noiseless amplifier, a series-feedback RIAA-equalized stage would have a signal-to-noise ratio (SNR) of 72 dB referred to 2 mV input at 1 kHz. This was calculated on the basis of a 600 mH moving magnet (MM) cartridge source impedance, and a 50k cartridge load resistance, both numbers I think well enough representative. This theoretical number could clearly be worsened by the noise contribution from realizable amplifiers, but on the other hand, could not be bettered however close to noiseless an amplifier was.

HPW’s own three-transistor design (basically the Bailey circuit) measured at 70 dB, thus it was 2 dB away from the theoretical best performance. I understand that early low noise ICs designed for the RIAA job, such as the LM381 (1972) and the NE542 (1977?), could match this level of performance, and that the NE5534/2 got about as close to the theoretical as reasonably possible.

Self (WW 1983 October, p.31ff) quoted an SNR of 84 dB relative to 5 mV input at 1 kHz for his NE5534-based RIAA stage. This equates to 76 dB at 2 mV input. At first glance it looks as if he found a way to better the theoretical. But a look at the “fine print” showed that this was measured with a 1k resistive source, perhaps more like a microphone than an MM cartridge. With a 600 mH source, one would expect something close to, but just below the theoretical. A TL071 in the same circuit measured at 69 dB (61 dB referred to 2 mV), well short of what is needed for the RIAA job, although fine at “line” levels.

Anyway, assuming that HPW’s mathematical analysis was correct (and I am not aware that it has been challenged), then the NE5534/2 would appear to be easily adequate for the RIAA job in noise terms, allowing a close approach to the theoretical best SNR. Other – probably more costly - opamps that measure better with low resistive source impedances would not seem to offer any advantage in situ.

One justification for the use of a flat gain stage ahead of a series-feedback RIAA stage is that the frequency-dependent nature of the feedback means that the input impedance of the RIAA stage varies with frequency, affecting cartridge loading and possibly in turn its frequency response. The preceding flat gain stage eliminates this effect. But I have only seen this justification made in qualitative terms, with no quantitative data. Whether it is a material issue I don’t know. If the cartridge loading is determined mostly by the input loading resistor (often 47k), then the impedance of the amplifier input itself is likely to be an order of magnitude higher. Thus even quite large shifts in the latter would produce only minimal change in the impedance seen by the cartridge. And MM cartridges usually have a not-too-narrow acceptable loading range. Something I recall is that the Quad 33 disc input stage was said to be purely resistive within ±5 degrees up to 20 kHz, which suggests that input impedance variation could be kept to narrow limits even with a simple discrete series feedback arrangement. That neither Self nor Baxandall saw fit to use a flat gain stage ahead of the equalized stage for their respective NE5534-based RIAA preamplifiers is I think strong evidence that such was not advantageous, “neither decorative nor useful” one might say.

As an aside, Linsley Hood favoured use of the initial flat gain stage for a different reason, which was that it enabled the use of his preferred shunt feedback equalized stage, which could then be implemented at low impedance (not constrained by the need for a 47k input arm) and at sufficiently high signal level that its relative noisiness as compared with the series-feedback arrangement was not a material problem.


Cheers,

[Craig Sawyers]

I note that the circuit schematic shows a single +12V supply, with the input of the first amp biassed at 6.28V. Which voltage will appear across C3 and C4. Although the time constant is only 22ms, it will take a while before these capacitors reach a low leakage state.

However, most dangerous is the output capacitor C7, which can only charge up via the input to whatever it is connected to. So a humongous switch on thump is absolutely guaranteed. If the input R of the connected device is 100k the time constant is 100 seconds.

Also the cut frequencies are: Input CR 3.5Hz, and the CR's of C3 and C4 are 7.5Hz each. Now that might be deliberate in order to give a bit of cut at LF, but it is poorly controlled (certainly for the 7.5Hz ones) because of the wide tolerance of electrolytics.

I would say that there are better designs out there based on the 5534.

Craig

[Craig Sawyers]

I implemented as a spreadsheet a table which appears in the National Audio/Radio handbook, Appendix 6, page 15. That divides 25Hz to 20kHz into 10 bands. (This classic tome has been reprinted; I bought mine recently from Amazon)

This calculates SNR, flat and with RIAA for cartridge L and R, capacitive loading and load resistance. With a noiseless amp.

For 50k/sub pF load capacitance, L=600mH and R=1 ohms and 2mV (Walker's parameters) it gives:

Flat SNR 59.5dB, and with RIAA 72.0dB.

Which is exactly the same as Walker's manual calculation for RIAA SNR of five decades ago.

What is rather interesting is that if the noise is dominated by a high gain flat buffer, the effective SNR is degraded as compared with wrapping the RIAA around the input stage. It also says that having a high gain input stage, passive RIAA and then a second flat gain stage is also going to have a degraded SNR.

Craig

[Radio Wrangler]

Yup.

You can arrive at the same conclusion by applying Friis' equation several times at different frequencies as the RIAA network changes the gain distribution in the multi-stage variants.

You can then complicate this further by considering how the frequency-dependent impedances of the network change the presented impedance to the amplifier and wander away from the optimum noise figure region. OK, so Smith charts don't normally get used at audio, but think of 'noise circle' contours of noise figure.

Then for a third issue, consider the overload behaviour. At one time there was a fashion for quoting headroom figures and engineering to make the numbers more impressive (Cambridge Audio?). The single stage approach comes out quite well here, too.

Glad Hugh's maths has aged well over the almost half-century!

David

[TIMTAPE]

Late to this thread. I agree that for MM's the standard circuit should be fine. The significant "real world" performance is the preamp noise, while the disc is playing.

What I find important to reduce is preamp noise that can be heard, and measured, again while the source is playing. For example, slow speed, narrow track analog tape recordings such as cassette or VCR recordings generally have a very low output at the repro head. Often, listening indicates, and the spectrum analyzer shows preamp noise, especially in the high frequencies, even with the baseline recorded tape noise playing.

The masking effect is often not well understood. If the source completely masks the preamp noise, in every part of the audible spectrum, then the preamp noise can be disregarded, regardless of whether it can be heard with no disc playing. If it's masked, it's masked.

The same applies when the roles are reversed. If the wanted programme in a recording is completely masked by random broadband noise, there is nothing we can do to retrieve it out of the soup.

[mhennessy]

Originally Posted by bikerhifinut

heres the circuit.

As others suggest, that's not a great circuit. By all means breadboard it if you think you'll learn from the experience, but personally, I wouldn't consider it a contender for the final solution as published.

If doing so, I'd change the power supply to +/-15V. That means you can lose the potential divider at the input, and just return pin 3 of IC1 to ground via a resistor. You'll want a loading cap in parallel - perhaps start with 100pF... Personally, I like to put a resistor to ground on the input side of C2 - it eliminates pops and clicks if you disconnect the input (if I remove the headshell to inspect/clean the stylus, I don't bother turning the volume down - there's no noise at all, amazingly). Value not critical, but remember it's in parallel with whatever you put between pin 3 and ground, so pick something that gives you close to 47k net. I went for 68k on the cartridge side of the input cap and 150k on the op-amp side. The NE553x has high-ish bias currents, so the higher you go, the more DC offset you have at the output (which is rarely a serious problem, but be aware).

If you replace C3 with a short, the offset will be amplified. This might not be a problem with 30V rails, but again, just keep an eye on it.

Personally, I would divide the resistors around IC1 by a factor of 100. As a rule of thumb, resistors larger than about 2k are potentially contributing more noise than they should, depending on the surrounding impedances, so always look to see if they can be reduced. Against that, if they're too low, the op-amps might struggle to drive them. The NE553x is superbly good at that; many others display a rise in distortion when asked to drive impedances below about 2k. But you could easily make R4 1k2 and R3 100 ohms. In simple terms, thermal noise in a resistor is proportional to the root of resistance, so a change of a factor of 10 equates to 10dB less noise - the reality in a complete circuit with other noise sources will be different, of course, but it's a handy rule of thumb to give you an idea of the ballpark.

If keeping C3, it needs to be something like 1000uF with R3 set to 100R. That's not just to ensure that this part of the circuit has an unreasonably low -3dB point at the LF end; rather, electrolytic caps show a rise in distortion when they are being used as a filter.

You could do the same scaling exercise with the components around IC2. Obviously it includes C5 and C6 as well.

Without doing a simulation - so I could be wrong - it does look like this circuit might have rather more gain than is necessary. Just in very "hand-waving" terms (to trigger the mathematicians ), the maximum gain at LF of the Self circuit (ignoring the sub-sonic filter) is x313, or 50dB. Whereas it's more like 60dB for yours, as you say. Let's say then, again in very simple terms, that the 1kHz gain is 30dB for the Self circuit and 40dB for yours. As I say, the Self circuit could do with another 10dB, which yours has, but your circuit has the 22dB amplifier ahead of it, suggesting that the 1kHz gain will be 62dB. For a 5mV input, that's 6.3V out! And if that's 5mV RMS, then that's practically 18V peak to peak at the output, which clearly isn't going to happen with a single 12V rail! Even if it did, whatever follows it will be expecting something rather more softly spoken...

So if you want to try this, try drop IC1 initially - I suspect that the values surrounding IC2 will give enough gain. No idea how good the RIAA curve matching will be - as I say, I haven't simulated it - but if this is a learning exercise, then there's much fun to be had by building it and measuring it. And maybe even listening to it

Don't forget a 4p7 stability cap if you use an NE5534 for IC2, as the gain at HF falls to unity.

Originally Posted by bikerhifinut

And is a resistor on the output a good idea for HF stability? Most commercial preamps seem to put something like a 500R to 1k here.

Yes, definitely. It can be as low as 50 ohms or thereabouts, and I'd suggest no more than 1k. Also good is a resistor from the output side of C7 to ground to ensure that C7 charges so that there's no DC offset at the output that might cause a pop when you first switch the input selector to this. Anything will do; 10k to 22k is a good nominal range.

Happy experimenting!

Mark

[Radio Wrangler]

Having seen that circuit, I would say that there is evidence that it wasn't by an experienced low-noise designer.

First off, the 100k and 91k potential divider splitting the supply is designed to look like the trad 47k ohm load impedance for the cartridge and saves at least one whole resistor. However, it is not filtered and transports muck on the supply rail straight into the most sensitive input you have. Yes there is a shunt 220u on the rail, but there is no series impedance for it to work with.

The feedback impedances of the first stage are unnecessarily high and will contribute additional noise as a consequence.

The turn on thumps and the lack of a dc drain resistor at the tail end have already been mentioned. The DC drain resistor ensures C7 is biased the right way round... usually! you don't know the DC characteristic of whatever it's driving.

Again the RIAA network is scaled for an unnecessarily high impedance. It's not as damaging as the first one, but why not get it better and use another NE5534? It'll have the oomph to drive the lower noise network and have plenty left over to drive a cable... Oh, I'd stich a 100 Ohm or so resistor in series with the output. Opamps do not like driving loads which can be capacitive at RF. So it makes sense to put a bit of protection in. The disadvantage of the two-stage approach has eluded the designer. I suspect the LF351 was chosen as being cheaper than an NE5534. But having no second stage would be cheaper still and actually work better.

As the arthmetic above shows, he hasn't been looking after the headroom aspect.

Another hidden characteristic of the 5534 is that if it's driven into clipping and is on balanced positive and negative rails, it tends to clip cleanly and come out of clipping without any hang-up. This is important. Records have bangs pops and scratches which in the very short term are far louder than the wanted content. Expect RIAA stages to get clobbered. The good ones pick up where they left off almost immediately where they left off. The bad ones can be left with their output stuck neat a power rail for a while before they recover. This massively exaggerates pops and clicks.

There's a lot more to designing the archetypal good RIAA stage than just playing pin-the-talk-on-the-donkey with an opamp data book.

But this is audio, and so the temptation to try to do more and more is overwhelming. You have to know when to stop.

David

[paulsherwin]

Would you proper engineers care to explain the drawbacks of minimalist single chip designs like the RJM VSPS?

This is a very educational thread for someone like me, who understands the basics but not the esoteric stuff.

[mhennessy]

At first glance, it's missing a couple of things, but it looks like it ought to be quite reasonable.

You don't want to risk putting DC through the cartridge, so an AC input cap is needed. That ideally needs another resistor on the cartridge side to ground. As described above, ensure the two resistors either side of the input cap look like 47k.

It's also missing an input loading cap, but in fairness it's mentioned in the parts list. Of course, 100pF is a good starting value.

I'd say that 2u2 at the output is a bit on the low side. The parts list says "non polar", but if this means "non-polarised electrolytic", then I'd be tempted to up it to 47u. If he means "non-electrolytic", then that's OK in terms of LF capacitor distortion - it's probably that low to act as a rumble filter of sorts - obviously the turnover frequency will depend on the input impedance of the following stage, so it's a bit undefined. The PCB appears to be laid out for an axial type with long leads, so perhaps the latter is what he had in mind.

There's no cap in series with R2, so check the DC offset. It'll be lower if you use the FET input OPAx134, but that'll be a bit noisier, so I'd prefer to stick with the 553x.

I haven't simulated it, but his results looks pretty reasonable. Self being Self, his circuit will be closer (it's +/- 0.05dB IIRC), but I wouldn't claim to be able to hear the difference

Self's core RIAA amplifier isn't really much more complicated that this one - basically an extra HF pole after the op-amp (as the RIAA curve should continue falling at HF forever, but a series-feedback design like this can't fall below unity gain). The Self design will be quieter because of the lower impedances used in the feedback network, but those shown here could be scaled accordingly, or you could just copy Self's values.

MM phono stages don't really need to be much more complicated than this to get the job done. The only downside of the "all in one" approach is the complexity calculating the RIAA values as they interact with each other - it seems to me that many of the alternatives seek to avoid this, at the expense of more stages that bring other compromises.

[mhennessy]

For reference, here's Self's MM pre, cropped tight so hopefully is readable. This is followed by a sub-sonic filter, which is shown in the second attachment. This is almost identical to the version I built, which has been in the public domain for many years - including nearly 20 years on my site, with Doug's knowledge - so I don't think there's any copyright issues here. If anyone looks up my version, note the error around the feedback resistors, and the notes of the conversation I had with Doug about it (confusion caused by bad printing and compounded by Doug, surprisingly, but corrected when Self on Audio was released). I will correct my drawings one day...

The main difference is that resistors at the input. Dropping the 220k to 150k gets you to 47k. I put the lower of the two on the op-amp side of the input cap.

As I've mentioned a few times, this only has about 30dB of gain at 1kHz, and could do with about 10dB more. When he originally worked out those values - for his "Precision Preamp '96" - he included an attenuator on the CD player input. Having just dug out my copy of the project, I see he chose a massive 16dB for that, on the assumption that the nominal level from non-digital sources was around 150mV. All my tuners, cassette machines, etc, from that era are more like 500mV, and the average level of well-recorded CDs is perhaps 10-20dB lower than the peak level, so the jump in volume between sources is not all that great in practice. As I say, with another 10dB, records would definitely be closer. But its not bothered me enough over the last 20 years to do anything about it! In theory, it's a software fix for me, as my pre-amp features input trim. Well, almost. The code for that is about 95% finished - it's almost certainly just another 2 or 3 extra lines of assembly code to incorporate the chosen value into the calculation of the two bytes that are sent to the PGA2310 volume control IC, but after 20 years away from the code, it'll take several days to reminding myself how on earth it originally worked!

[Radio Wrangler]

This is easy, Paul!

There is very little wrong with that circuit. Mark's picked up the main points I've spotted.

The NE5534 type opamp is about as good as you get. If you want the last bit of precision/performance, there are a couple of possibly better opamps but the next step is to a full discrete design and you are far into the realm of diminishing returns. With my engineering hat on, this is as far as I feel the need to go. To a true audiophile nothing short of the ultimate will ever do, and just knowing that it could be pushed further is enough to spoil their day.

The input... The Ne5534 class of opamp have about a microamp of bias current flowing in their input pins. Those two inputs go straight into the bases of two transistors. Those transistors are optimised for low noise and good bandwidth. Not for DC gain, not for DC precision. So compared to many opamps the bias current and offset voltage are not very good. And it doesn't matter in this application.

It's not good practice to have DC current flowing into the coils of a magnetic transducer like a cartridge. The world won't end at this current, but a series capacitor on the input would block it and make you feel a bit more comfortable. That sort of current into a tape head, on the other hand would be trouble, it would magnetise the head a bit.

The RIAA stage has max gain at low frequencies and this one has plenty.

I would be inclined to scale the value of the added input capacitor to do a bit of rumble filtering because any rumble is going to meet a lot of gain. So with the input resistor at 47k, I'd opt for 10 or 15Hz as the -3dB point, so a capacitor of 330nF non electrolytic would be fine.

That roughly 1uA bias current (DC) in each input will flow through R1 and R2 (we can neglect R4 in a back of a fag packet calculation) these resistors are quite different 47k and 680R so we will generate a voltage difference of 1uA*(47k -680) which is 46.3mV effectively at the input. U1 has a lot of DC gain (R4/R2)+1 which is 1130 so that offset voltage, plus the opamps own offset (about 5mV) gets amplified so the output will sit away from the centre by a fair amount. Fitting the capacitor to block DC in the cartridge has made things worse for our amplifier It will try to create an output offset of 56 volts ! Jeepers I've wrecked it!

But with the cartridge shorting the input without that added capacitor I'm still going to get an appreciable offset. This is the main oversight in this simple circuit, but it's easily fixed. We put a DC block in series with R2. It rolls the amplifier gain down to 1 at DC so things get 1130 times better. Let's shoot for 10Hz again, and we get a bit more rumble filtering as well 680R and 10Hz comes out as 23uF. There is no reliable DC bias voltage on this, so that rules an electrolytic out though some tants run OK at zero bias. You don't need many volts working, but a plastic film capacitor of some sort is going to be costly.

So what we've found out is that to avoid a pricy capacitor, the circuit runs a bit of DC through the cartridge and may run the opamp with the output a few volts off centre at its output.

This issue is made a bit worse by it being quite high gain as such circuits go. I'd be tempted to scale down R4 and R3 and scale up C1 and C2. These values will reduce the gain and be a bit less noisy. The reduced DC gain from R4 will help with the DC offset.

So will it work? Yes, quite well as it stands.

Can it be improved? Yes, There's a bit of bad practice, fixing which means chasing some knock-on effects. And I think it's too gainy in the rumble frequency range.

As said above, a capacitor to ground shunting the input is needed to give the recommended loading for many cartridges (don't forget to allow for the C of your cables)


The provision of R5 is good. Someone knows that opamps can go unstable into capacitive loads directly on their output, like a length of coax.... And he understands that the non-inverting style has a noise advantage.

A purist would say that the non-inverting circuit makes the RIAA roll off floor at unity gain, when it should keep going down. This is a smaller issue the higher the gain of the stage is set for. We're talking about maybe 1dB at 20kHz. Can't say I'm bothered too much, but Peter Baxandall, no less, published a circuit very like this one with an added capacitor to fix this error in the letters column of WW.

David