Measuring noise and SINAD of MM phono preamps properly

[Michael Fidler]

NJM8068 and NJM8080 are the new versions of the NJM2068 and NJM4580 respectively. While they're a little bit faster on the test bench, I wouldn't want to use them in a conventional input/equaliser/single-stage phono preamplifier as all the new stuff that Nisshinbo are making seems to suffer from excess distortion once the closed loop gain exceeds 30dB or so. This is fine for conventional line-level work, but if you're building an RIAA equaliser, or a standard series-feedback MM front end/equaliser then you'll see the distortion below 300Hz really kick up: about a five-fold increase for the 80 series or anything with a "C" at the end of the part number (i.e. NJM5532C or NL8802).

You can still use them for phono front ends (they are very good on voltage/current noise for MM impedance, perhaps the best currently available despite the lower cost in comparison to TI/Burr-Brown devices), but the topology has to be modified so the closed-loop gain stays below 20dB, then hands over to something like a TI SA5532 for the low-frequency gain/equalisation.

 

[ryanmh1]

I have the Cambridge now. Didn't trace it out yet, but it's probably not possible to tell whether it's a fet or bipolar. The SMD code leaves too many options.

There are two of these circuits, basically identical, for both MM and MC. Same transistors in each? Who knows. If any older products have service manuals, they may be similar.

 

[Heinrich]

I have the Cambridge now.

Congrats, and I would assume that these guys optimized the two stereo inputs acordingly. But there should be three, actually.

But first, what's for, this engineering, the target? Of course we all come from an absolutist's camp. We want it all and more, a truckload more, as they say. Because with phono that is relatively cheap. Think of equivalent scruitiny put into power amps or speaker boxes. Those would be huge and quite expensive at multiple tens of thousends, while no phono pre manufacturer I know of charges more that 2k or so. One could enjoy having the top notch solution at least for a singled out part, and forget about all the other stuff deliberately. That's human--focus is exclusive, by the word.

From a cold hearted engineering perspective things are way easier.

1. actually point (0), a disclaimer: phono is not limited, but defect ref/ contemporary standards; we're polishing BS anyway
2. a few roughly equal weighted noise sources are to be considered
3. the LP itself from material cutting the matrix, over lossy pressing the actual copy to multiple playback in dirty environment
4. the pickup's self noise (Johnson) MC << MI < MM
5. voltage noise of the pre's input stage
6. current noise of the pre's input stage mirrored at the pickup's impedance
7. maybe internals of the pre amp, e/g passice/active RIAA shaping

WIth modern, kind of standardized methods, we measure just and only (5). In order to measure (6) also we would need to know (4) as to substract this after reading. (That's where apparently most of the vinyl fellows just quit.) Point (3) makes me ask, why (4..6) matter that much--if the little hiss when changing records isn't bearable, while during the playback at least double, or quadruple noise is. (Some players shorten output with tonearm in rest position.)

This brings me to my argument. See post #135, again. It is a competent (!) decision to accept higher numbers on (6) for versatility at lower cost. Point (4) is three sided. For MC the benefit is clear. For MI (moving iron) the same benefit is gained, but MI is not that common (Grado). For MM the compromise shows: a noise penalty of about 6dB (hiss). But, the resulting numbers are still perfectly bearable considering (3) and regular use cases, especially under (1). For instance the Fosi X5 used w/ MM is as good as many mid tier offerings at lower or equal cost. Just roll the numbers, if you can. Or practically simple, shocker, happily crank your system up as to hear that hiss, and then plopp the needle to the groove - still that daring? See (1).

Times evolve, tables are turned. With digital as a comparison the vinyl analog is a drag. In hindsight much of the effort taken was running in a hamster's wheel--for big money getting nowhere. One might want to perpetuate that psychological stance out of sentiment. Then take three: one for MC, one for MI, and another one for MM and pay for your choices.

 

[ryanmh1]

phono that is relatively cheap. Think of equivalent scruitiny put into power amps or speaker boxes. Those would be huge and quite expensive at multiple tens of thousends, while no phono pre manufacturer I know of charges more that 2k or so. One could enjoy having the top notch solution at least for a singled out part, and

[...]

4. the pickup's self noise (Johnson) MC << MI < MM
5. voltage noise of the pre's input stage
6. current noise of the pre's input stage mirrored at the pickup's impedance

WIth modern, kind of standardized methods, we measure just and only (5). In order to measure (6) also we would need to know (4) as to substract this after reading. (That's where apparently most of the vinyl fellows just quit.)

Times evolve, tables are turned. With digital as a comparison the vinyl analog is a drag. In hindsight much of the effort taken was running in a hamster's wheel..

Apologies in advance for the wall of text. This turned into a lot more than I thought it would be at first. I also realized I ignored your points, to a degree. More or less, the point there is that the differences in #4 don't matter enough to matter for most MM carts because the Standard Load discussed below will get within a dB or two for virtually all of them, which also takes out #5 and #6. You hit the numbers below, you've solved the "problem" of transparency where noise is concerned in all cases with a MM pickup. I think. So, away we go...

I'm not sure what price has to do with it, and you're well off the mark there on price. Plenty of phono stages cost well more than $2000. I would wager a guess that there are probably more over that price than under--not in terms of sales, but different models and makes. Plenty of speakers costing many thousands also.

With phono stages, just belittling the format and saying, "It stinks so who cares" is unhelpful. The debate is not whether the format is the best, but whether the engineering of the products to reproduce it are good, and making sure you aren't making worse something that's already arguably marginal.

The base assumption is that any phono stage under around, oh, $300 to $400 is far more likely than not to be used with a moving magnet cartridge. These days, probably an Audio Technica or Ortofon. Thus, it makes sense to measure that (particularly when measuring at 40dB gain) with a "standard" cartridge load that reasonably approximate it. Is that "standard cartridge" going to perfectly represent each cartridge? No. However, it is a provable (and proven) fact that a measurement of SINAD of, say, 76dB (5mv, unweighted) can drop as low as, say, 60dB with cartridge connected. That may not be typical but it's possible. That's well below the best 33 1/3 pressings, which are around -72.5dBA (Vogel, Sound of Silence, p139), ref 8cm/sec. For a 45, it's -75.5dBA. Vogel then shows that a "worse case" 45 requires over -77dBA ref 5mV to avoid the phono pre degrading the record. That also happens to be about the same as the "best case" 33 1/3. Vogel uses dBA for an arguably good set of reasons that aren't worth getting into at this point too deeply (since I'm sure that discussion has already happened), but suffice to say -85dBa or so is probably a good "guaranteed transparent" target, cartridge connected. For average records, this is guaranteed to be fine, and limiting to 1/4dB degradation only requires about -82dBA.

From here, we can use some of the other tables that Vogel supplies to make a few educated guesses and inferences. A Lehmann Black Cube (Table 3.11, p142) measured at -75dBA with Standard Load (1k+500mH+125pF+47k5). Lehmann specifies it as -71dB unweighted, ref. god-only-knows. ASR measured a Lehmann Decade unweighted as -78dB unweighted, which is exactly where it is specified. Since it's a good manufacturer, we'll assume the same math applies when attaching Standard Load and A-weighting (a bold assumption). More or less, we subtract 4dBA from the Black Cube's number, or -82dBA. That's probably transparent in terms of noise. Just.

We also happen to know that something like a Fosi X5 goes from about 28uVrms noise to 245uVrms noise with cartridge connected. So, connecting that cartridge drops us from, say, -85dB to -67dB. From there, you have to try to apply an A-weighting filter, which I took a best guess at by binning the noise floor measured here a year ago by @nagster. You were involved in that discussion. Based on the spectrum of the noise floor, I'm not optimistic it's going to "gain" any more than an extra -10dBA. More like -7dBa. Depends whether you include or cut off PSU noise. So my rough guess is with a cartridge, you really get around -75dBA out of the Fosi, referenced to a 5mV input on a Standard Load. But look what just happened there. The OPA1612 Fosi just dropped well behind the Lehmann, and on par (dBA) with a positively ancient Lehmann Black Cube with an SNR spec of 71dB. Hmm... There is a disturbance in the force.

Plus we know the Fosi's OPA1612 has worse MM noise than an NE5534A by a good bit. Finally, we know that we can improve on that NE5534A by about 5dB with the best discreet implementations, such as the X-Altra from Hifisonix. So, anything using a single OPA1612 for an input opamp cannot be anywhere close to "state of the art."

Summary: Assuming all of the very rough math and conversions work out right, 85dB unweighted into a shorted load is not guaranteed to be below the noise floor of the best vinyl records. It's also very difficult to squeeze out those last few bits of hypothetically possible performance.

So, just how much these products degrade with a Standard Load connected is, in the end, possibly important to not degrading what is on the disc. A-weighting that result provides a further solid reference regarding the audibility of the noise spectrum, but that's slightly beyond the scope.

All of this does answer one other fun little question: Why do audiophiles like vinyl? Well, I wouldn't be surprised if a lot of this actually was borderline audible. I don't recall where I derived that threshold as being when I looked at that last year, but my recollection is that it was somewhere in the high 70s or low 80s. So that's way more "fun" than DACs and properly implemented solid state amplifiers. There is literally no point in reviewing most DACs. It's all more than good enough and at "appliance" level. Same goes for many power amplifiers. But I am no longer going to make fun of someone when they say "cartridge A" sounds different on "phono preamp A" or "phono preamp B" because it's quite possible they actually might. Mind blown.

(That said, because we're at the threshold, it shouldn't be all that hard to definitively push past it for a couple hundred bucks if a designer knows his stuff, which so far few have, [EDIT:] and which the measurements in this forum in any event are incapable of revealing. Had to add in that last little bit, since I think it is so important on a forum dedicated in some part to proving subjectivists wrong. ASR measurements alone never will, which I hope is anathema enough to Amir's ears to implement some "best practice" changes. Namely, implementing A-weighting, Standard Load, and possibly 19+20kHz IMD tests in phono reviews. Do I actually expect that to happen? Probably not, but we can hope. How much more difficult and complicated this all was than in digital audio--and actually does all operate right at that threshold of audibility/transparency--I had no idea, and I doubt neither did he. That's not an indictment by any means. But you just can't prove a favorite ASR punching bag like Michael Fremer wrong when you do a bunch of acoustics and analog electronics/phono research, run the maths, and determine he quite possibly isn't. At least, not about one thing. Everything else like cables and most adequately powered, competently designed solid state amplifiers? Still wrong.)

 

[Michael Fidler]

All of this does answer one other fun little question: Why do audiophiles like vinyl? Well, I wouldn't be surprised if a lot of this actually was borderline audible. I don't recall where I derived that threshold as being when I looked at that last year, but my recollection is that it was somewhere in the 70s. So that's way more "fun" than DACs and properly implemented solid state amplifiers. There is literally no point in reviewing most DACs. It's all more than good enough and at "appliance" level. Same goes for many power amplifiers. But I am no longer going to make fun of someone when they say "cartridge A" sounds different on "phono preamp A" or "phono preamp B" because it's quite possible they actually might. Mind blown. (That said, because we're at the threshold, it shouldn't be all that hard to definitively push past it for a couple hundred bucks if a designer knows his stuff, which so far few have.)

Yes, in a way it's more satisfying because it's much more difficult, and spending money can lead to obvious improvements as the limitations are so severe!

 

[Heinrich]

The base assumption is that any phono stage under around, oh, $300 to $400 is far more likely than not to be used with a moving magnet cartridge.

I'm happy to own a Denon DL103, bought already out of sentiment some 20y ago. Now reviving the vinyl zombie. It works!

These days, probably an Audio Technica or Ortofon.

Take Grado, because it benefits from the Fosi's design perfectly, as it is a low inductance MI.

Vogel then shows that a "worse case" 45 requires over -77dBA ref 5mV to avoid the phono pre degrading the record.

I don't know Vogel. But what follows is a typical focused view with logically limited perspective. It is not the least practical. Who in the world would dare to amplify a regular phono stage to perceptible hiss, and then drop the neede to the groove? Nearly all speakers would shatter to pieces caused by the BUMM/BANG - with the Fosi even the subsonic is "missing", as some put it. And does Vogel live in a dust-free environment, playing their record only once? They get the very first pressing of hundreds to thousends of copies exclusively? You, as I feel it, ignore 9x% of reality when following Vogel.

Not the least, you missed the point of self-noise of an MM pickup entirely. It limits the total noise in the chain of those humbly contributing little guys. (I pointed to a spreadsheet covering the topic, you didn't look it up, o/k.)

Standard Load, and possibly 19+20kHz IMD tests in phono reviews.

Raises the question, did you ever measure the distortion inherent to the mechanical composition of vinyl?! I did .. LOL. Let's get real.

 

[ryanmh1]

I don't know Vogel.

Not the least, you missed the point of self-noise of an MM pickup entirely. It limits the total noise in the chain of those humbly contributing little guys. (I pointed to a spreadsheet covering the topic, you didn't look it up, o/k.)

Burkhard Vogel's entire book is largely math and formulas. Exploring vinyl and where the limits are with cartridge connected. Did not miss the self-noise. That's going to be around -75, if memory serves. Noise is additive, so the self noise isn't all that material. -82dBA remains a good target. And you can just use a HOMC to obviate that issue...

I'm sensing a lot of changing the subject from you instead of dealing with the validity and analysis of technical points. Honestly, it's fatiguing. You know the darn math, but it's always, "Well, this other aspect of vinyl sucks or vinyl just sticks in general so who cares."

It ain't digital where we're 25dB past "who cares" into digital black. This stuff actually still matters because analog is hard, and is often gotten wrong here in spades, and has been for years. It's well past time to clean it up. People are wandering around believing their $80 boxes are perfectly transparent, when it's provable that on the best pressings, they're likely not.

 

[Heinrich]

Thanks for clarification. I don’t come from a mathematics background. For example, the question of what is worse: 3% distortion, or 10% at level peaks. At some point it doesn’t really matter—you have to learn to listen past it anyway and actively focus on the musical content. I think that’s a good, albeit basic, question, and I like the answer as well. This mental engagement is what makes these music boxes appealing to me.

As for noise, I would argue for using the concept of noise figure. That requires separating the noise contribution of the source from that of the subsequent amplification and playback. If we start on the playback side, we have ambient noise—wait, yes: I mean the medium is so poor (by modern standards) that really all aspects should be taken into account to justify its use, since it is no longer the only option, logically speaking.

Ambient noises such as birdsong in a north-western summer, your own breathing, and so on are in the 20 to 30 dB range. The question now is how loud it should get at peak modulation. From that, the required signal-to-noise ratio relative to the 20–30 dB of ambient noise can be determined. And don’t forget: the pop when the needle drops into the groove will always be well above the peak level, and it puts a corresponding strain on ears and loudspeakers.

From the other side, a more realistic approach—one I insist on—is to estimate the signal-to-noise ratio of ordinary records after years of use. And yes, anyone who still buys brand-new vinyl today is completely outside the scope of my consideration. I would say that 60 dB is a good value. I have records that probably reach that level, and I also think that analog tape plays a role here, because the noise “breathes.” In any case, I do hear groove echoes now and then, not to mention the occasional clicks and pops.

That brings me to a new idea: it shouldn’t get too loud at all, because any disturbances become disproportionately more annoying the louder everything is—and we’re talking about very relaxed hi-fi here, don't we?

But that has taken us quite a bit away from the topic. As I’ve said many times, I think SINAD simply doesn’t make sense for phono. The medium itself must be considered as an ultimately insurmountable limit—especially when it comes to distortions (SINAD!). As you asked for difference tone distortion, here you get it:
5000/5200 and 8000/8400 @7cm/s or so, well within full scale
doesn't need no maths, this is plain horrible by all technical standards 10% distortion and more and more and it is all in the vinyl's mechanical design.

You want (somebody else!) to cross check every phono RIAA amp if they match, doubt the competence of engineers, fear a lack of transparence, transparent to what?

My conclusion: if one doesn't embrace the 'limitations', the magic is amiss

 

[EJ3]

And don’t forget: the pop when the needle drops into the groove will always be well above the peak level, and it puts a corresponding strain on ears and loudspeakers.

My Technics SL-M3 does not do that, as the electrical to the cart is not turned on until the needle is on the record.
How it does that, I do not know.
These TT's were manufactured from 1984-1988, I believe.
It's something that I have forgotten about, as I have had it a such a long, long time that I don't notice that the needle drop noise is not there.
Many, many years of mental acceptance of a noise that is not there has become normal for me, I guess.

 

[markstein]

My Technics SL-M3 does not do that, as the electrical to the cart is not turned on until the needle is on the record.
How it does that, I do not know.
These TT's were manufactured from 1984-1988, I believe.
It's something that I have forgotten about, as I have had it a such a long, long time that I don't notice that the needle drop noise is not there.
Many, many years of mental acceptance of a noise that is not there has become normal for me, I guess.

My needle drop is audible but does not cause clipping at the Waxwing input @62dB gain, so - no issue if you don't crash the arm onto the record IME (equipment see signature).

 

[ryanmh1]

as an ultimately insurmountable limit—especially when it comes to distortions (SINAD!). As you asked for difference tone distortion, here you get it:
5000/5200 and 8000/8400 @7cm/s or so, well within full scale
doesn't need no maths, this is plain horrible by all technical standards 10% distortion and more and more and it is all in the vinyl's mechanical design.

It may be inherent to the medium or a cartridge, but since there are so many of those with varying production processes, it's difficult to control for it. All we can control for are the electronics. My thinking with the IMD test is that the RIAA curve is spitting out high frequencies from the record at a fairly high level due to pre-emphasis, and it's quite possible that some designs aren't going to have adequate feedback to control for it. A perfect example of this is the Darlington Labs MP-7, which when blipped only enough signal to output 1V will respond by spraying out second order products at a whopping 3%. On a 1kHz tone, it will only put out .1%. https://www.soundstagenetwork.com/i...p-amplifier-measurements&catid=414&Itemid=547 So what looks like a "good enough" design for vinyl on a simple test is an absolute mess and not nearly "good enough" when the going gets a little tough. I assume their MM-5 does the same thing since it's the same set of design principles. The ASR tests did not capture it. Hence, the need for that test as well.

Small backstory here, but years (decades) ago I tested probably some four dozen FM tuners, AFAIK the first to do with with an FFT. The big challenge was dealing with a "strange" product that had all sort of things going on from adjustable filters, demultiplexers, distortion cancellation circuits, pilot tones, etc. It was interesting how once you hit on the "right" set of tests, they actually aligned pretty darn will with subjective impressions. Phono stages, by comparison, are a piece of cake. Still, that's something you can actually pull off with some validity with analog sources, whereas with most digital sources, you're essentially claiming to hear the sound of a lightbulb illuminating.

Overall, there are a lot of stupid design decisions in phono preamps, and you need more than SINAD and overload to sort them out, IMHO. You really do need SNR with cartridge connected, CCIF IMD, and A-weighted cartridge connected SNR to sort it all out. There may be others. Also, the A-weighted measurements have been shown by a variety of bright individuals to correspond to audibility of noise. Hence, you get something like the Audio Technica AT-PEQ30 which has a stunning 35dB difference between A-weighed and unweighted SNR. https://www.soundstagenetwork.com/i...reamplifier-measurements&catid=414&Itemid=547. Despite being not great in absolute terms, it may "sound" extremely quiet after weighting. That's an extreme example, but since we're dealing with audible noise in any event, the shape of that noise floor probably will matter for how a product sounds, and A-weighting captures it fairly nicely.

When people say, "this thing sounds noisy" but it "measures" on whatever test your are using as being "better" than a "noisier" product, the problem probably is the test being utilized, not the ears hearing the noise.

 

[Heinrich]

... https://www.soundstagenetwork.com/i...p-amplifier-measurements&catid=414&Itemid=547 So what looks like a "good enough" design for vinyl on a simple test is an absolute mess ...

The point goes to you, thank you for that! I understand it like this: the RIAA equalization reduces distortion damping toward sub-harmonic frequencies. The argument is very strong — I hadn’t thought about it at all. See me corrected.

Overall, there are a lot of stupid design decisions in phono preamps, ...

Apparently.

You really do need SNR with cartridge connected, CCIF IMD, and A-weighted cartridge connected SNR to sort it all out.

I lately provided amateurish measurements comparing MM versus MI (moving iron, low inductance) at a Fosi X5. (btw, the design is not prone to above mentioned, ehm, difficulties.) Response wasn't responsive, though. Seems open source isn't in demand.

When people say, "this thing sounds noisy" but it "measures" on whatever test your are using as being "better" than a "noisier" product, the problem probably is the test being utilized, not the ears hearing the noise.

Listen to people, I'm supporting that.

Not particularly for you (me like those really critical listeners!):

Stereo Lab - Manufacture of gramophone records

Stereo Lab Manufacture of gramophone records FRANCINSTIEN Stereo Lab FRANCINSTIEN T-Sym NEEDLE=DROP NEEDLE DROP software equalisation recording charateristics quadraphonic quadraphonics ambisonic ambisonics SQ QS quadraphonics quad binaural stereo Blumlein cassette noise reduction in software...

pspatialaudio.com

plus (a bit optimistic, though)

Stereo Lab - How do LP records measure up?

Stereo Lab Stereo Lab - How do LP records measure up? Are records better than CDs? Alastair MacMaster Richard Brice music electronics music engineering PHAEDRUS AUDIO

pspatialaudio.com

 

[morillon]

I imagine this old document has already been mentioned in these discussions...(?) ;-)

 

[AnalogSteph]

NJM8068 and NJM8080 are the new versions of the NJM2068 and NJM4580 respectively. While they're a little bit faster on the test bench, I wouldn't want to use them in a conventional input/equaliser/single-stage phono preamplifier as all the new stuff that Nisshinbo are making seems to suffer from excess distortion once the closed loop gain exceeds 30dB or so. This is fine for conventional line-level work, but if you're building an RIAA equaliser, or a standard series-feedback MM front end/equaliser then you'll see the distortion below 300Hz really kick up: about a five-fold increase for the 80 series or anything with a "C" at the end of the part number (i.e. NJM5532C or NL8802).

Very interesting, I had no idea that performance would be that different. (How much is it in absolute terms at say 100 Hz in a 1-opamp phonopre, and what order, and how does vinyl compare?) Any clues as to why that might be? I would think that process improvements would have permitted reducing current levels a fair bit while still allowing for somehat increased GBW, which may be backfiring. (This seems more likely than reduced OLG, which would have shown up in specs.) It should show in load immunity testing then. It is known that using an NJM8065 as a headphone amplifier of near-zero output impedance does not go over particularly well, as evidenced by Focusrite Scarlett 3rd gen performance (supplies either +/-5V or perhaps +5V only):

If it is not an OLG problem, a little dynamic microphone preamp with ca. 50 dB flat gain (not a rarity on soundcards about a quarter century ago) should make a good test case. It may be fun to introduce an NJM2122 as a "secret weapon" for grins, but do observe its voltage limits (+/-7 V for the M and +/-10 V for the D).

 

[Michael Fidler]

Very interesting, I had no idea that performance would be that different. (How much is it in absolute terms at say 100 Hz in a 1-opamp phonopre, and what order, and how does vinyl compare?) Any clues as to why that might be? I would think that process improvements would have permitted reducing current levels a fair bit while still allowing for somehat increased GBW, which may be backfiring. (This seems more likely than reduced OLG, which would have shown up in specs.) It should show in load immunity testing then. It is known that using an NJM8065 as a headphone amplifier of near-zero output impedance does not go over particularly well, as evidenced by Focusrite Scarlett 3rd gen performance (supplies either +/-5V or perhaps +5V only):

If it is not an OLG problem, a little dynamic microphone preamp with ca. 50 dB flat gain (not a rarity on soundcards about a quarter century ago) should make a good test case. It may be fun to introduce an NJM2122 as a "secret weapon" for grins, but do observe its voltage limits (+/-7 V for the M and +/-10 V for the D).

In absolute terms, the increase in high-closed-loop-gain distortion is about five to tenfold at 100Hz/54dB gain in a typical 40dB RIAA amplifier, with the products being predominantly second and third harmonic. For instance, the NJM2068 in the Spartan 15 produces roughly 0.0005% THD at 100Hz, maximum output, whereas in the Spartan 16 prototype from the beginning of 2025 I measured 0.004% using the NJM8068 on the front end. The NJM8068 does seem to produce slightly lower HF distortion than the 2068, though. I have it in my notes, but not with any concrete figures, I'm afraid. The next time I get my old laptop out, I'll see if I have any traces that demonstrate this.

I don't think the increased distortion at high gain has much to do with GBWP, rather I would wager over the odds that it's due to measures taken against RFI on the input stage of the IC; likely some form of degeneration or cross coupling that limits the LTP in correcting nonlinearities in the following VAS/mirror stages. While NJM/Nisshinbo are notoriously vague with the internal of their devices (like most IC manufacturers), and if my memory serves me correctly, a proximal insider who has the jump on new their new stuff informed me that the 80 series and C-suffixed devices were being made using an improved foundry process for increased temperature performance and radio-frequency rejection.

Regrettably, with the ubiquity of direct-from-China switching electronics that bypass typical EMC requirements in the domestic environment, expected levels of audio-detectable RF are much higher than the legislation would imply. Increased filtering and HF-strapping of PN junctions are becoming obligatory for manufacturers who don't want to get the the usual complaints of "buzzes under my LED lights".

In relation to the NJM2122: this is quite a tricky IC as it's not stable at closed-loop gains below 30dB, but there are workarounds, as you can see in the sketch below for a low-cost LOMC head for an MC-PRO-style topology, yielding a noise figure of 12dB for a 10 ohm source with low-cost SMT components. The key is to use it as part of a nested feedback loop with a more drive-and-voltage-competent IC, the 5532 being the prime candidate. By wrapping in the 5532, you get an output stage that can drive 10V RMS into the next stage (in this case a shunt-feedback equaliser), while also giving yourself a handy insertion point for a frequency-compensation network (R5, R6 and C2 for the left channel). As the 2122 isn't driving a heavy load (not that it can due to the simplistic class-A output stage), you can bring the voltage down with a highly-reliable/low-cost zener shunt regulator.

I never actually built this circuit, as while the 75dB SNR for a 0.5mV cartridge would be considered competitive for the market on account of fraudiophilia's dysfunctional technical performance in comparison, making a low-cost LOMC stage is a highly unattractive prospect on the customer service side!

 

[AnalogSteph]

In absolute terms, the increase in high-closed-loop-gain distortion is about five to tenfold at 100Hz/54dB gain in a typical 40dB RIAA amplifier, with the products being predominantly second and third harmonic. For instance, the NJM2068 in the Spartan 15 produces roughly 0.0005% THD at 100Hz, maximum output, whereas in the Spartan 16 prototype from the beginning of 2025 I measured 0.004% using the NJM8068 on the front end.

Meanwhile, a good phono cartridge will make it to perhaps 0.3%. So as I suspected, it's a bummer in terms of measurements but ultimately you still have two orders of magnitude worth of margin instead of three. Even relatively crude 1970s circuitry (with maybe one order of magnitude to spare) can do the job quite acceptably.

In relation to the NJM2122: this is quite a tricky IC as it's not stable at closed-loop gains below 30dB,

Indeed. I always wonder how inexpensive audio interfaces are dealing with that, as the chip is ubiquitous in their variable-gain microphone inputs which can clearly handle substantial levels (like +9.5 dBu for a Focusrite Scarlett 2nd/3rd gen). I see nothing particularly special in the circuitry of the old 1st-gen 2i4 (pot presumably 10k-50k, perhaps 20k):

So clearly they are letting the 2122 swing as close to the +/-5 V rails as possible and are using some passive attenuation for level matching and are using some small caps across the 10k feedback to keep the peaking at bay, but that's about it. That'll do?

 

[Michael Fidler]

Meanwhile, a good phono cartridge will make it to perhaps 0.3%. So as I suspected, it's a bummer in terms of measurements but ultimately you still have two orders of magnitude worth of margin instead of three. Even relatively crude 1970s circuitry (with maybe one order of magnitude to spare) can do the job quite acceptably.

You're almost certainly right there. Especially at the 20mV or so of cartridge that gives 10V RMS line output from a typical 40dB phono preamp.

Indeed. I always wonder how inexpensive audio interfaces are dealing with that, as the chip is ubiquitous in their variable-gain microphone inputs which can clearly handle substantial levels (like +9.5 dBu for a Focusrite Scarlett 2nd/3rd gen). I see nothing particularly special in the circuitry of the old 1st-gen 2i4 (pot presumably 10k-50k, perhaps 20k):

That is quite an interesting image, if drawn correctly. Seeming to imply something else is going on as that's a hell of a lot more feedback than the -30dB required for stability with the 2122. That might warrant a bit of experimentation, although my suspicion is that the part is misnumbered?

So clearly they are letting the 2122 swing as close to the +/-5 V rails as possible and are using some passive attenuation for level matching and are using some small caps across the 10k feedback to keep the peaking at bay, but that's about it. That'll do?

Yes, a classic "instrumentation amplifier" config. Passive attenuation might break the don't-attenuate-and-amplify adage, but perhaps for such a low-noise device it's acceptable.

 

[ryanmh1]

I have the Cambridge now. Didn't trace it out yet, but it's probably not possible to tell whether it's a fet or bipolar. The SMD code leaves too many options.

Following up on this, Cambridge used to publish service manuals, which I've located now. They have basically used the same design for some time, and there is little doubt this is the same input stage design as those older published version (based on the layout and parts count), but updated for SMD parts. They have not changed the design in 20 years. It's a discrete differential design using three bipolar/BJT 2SA970 transistors in parallel, so 6 total per channel. Presumably, that would now be the SMD equivalent. Schematic is here: https://audiocircuit.dk/downloads/cambridge/Cambridge-640P-riaa-sm.pdf. Would have been nice if it was FET based, but oh well. Given how well it measured with input shorted,, I suspect it was am not, and it ain't. So, yet another one that probably does not warrant its place atop "the chart". How bad it gets is harder to figure since it's all discreet and depends on other factors I'm not ambitious enough to attempt to calculate.

EDIT: Forgot to mention it's not unreasonable to suspect Doug Self played some role in some or another part of this thing's design. He's done some work for Cambridge before, and his descriptions of his design in Small Signal Amplifier design bears some obvious similarities, like selecting "three paralleled transistor for cost reasons." Hmm... I suspect someone over there went, "The MC section looks great! Use it again for the MM amp and change the gain!"

To relate this back to the original subject of measurements, the fact of the matter is that the whole "chart", above about 74dB or so, can fairly be said to be wholly unrepresentative of anything useful whatsoever, except for engineering stupidity by many designers. Ye olde "chart" does not even represent that which it claims to represent in the first place for virtually anything on it--the quietest moving magnet phono preamp. Oh, well. Frankly, I would not be surprised if the best one left were the Schiit Skoll (which is probably JFET and an otherwise stupid no/low feedback design), the Lehmann, obviously the Spartan 20, and possibly some other cheap pieces of junk that used an NJM2068 or NE5532A by happenstance.

The real issue I take with this is that it would at first seem like you could get something with SOTA performance on a beer budget. But nothing measured yet shows that at all (nor will it ever absent methodology changes), which sort of defeats much of the point. Truth is, I suspect the "beer budget" winner would be the Aiyima T3 Pro if it has sufficient overload margins, since it has a swappable input opamp, includes some sort of low pass filter, and hypothetically comes with an NJM2068 out of the box, and the manufacturer measurements look promising. But it definitely starts showing distortion at 40mV and completely overloads by 100mV (ascertainable from about 1/4 second of their "production process" video with validation testing, so unless it's using active RIAA, overload could be questionable.

. For instance, the NJM2068 in the Spartan 15 produces roughly 0.0005% THD at 100Hz, maximum output, whereas in the Spartan 16 prototype from the beginning of 2025 I measured 0.004% using the NJM8068 on the front end. The NJM8068 does seem to produce slightly lower HF distortion than the 2068, though. I have it in my notes, but not with any concrete figures, I'm afraid. The next time I get my old laptop out, I'll see if I have any traces that demonstrate this.

Point 2 is how much better the Spartan 20 actually is due to your measuring for appropriate parts than perhaps everything else to date is, and by a significant margin, but gets no credit for it. It is not just slightly better than something like a Fosi or the Mani 2 (also OPA1612) and probably even the Cambridge. It's a whole lot better, when you consider the Fosi will drop down a whopping 18dB or so as measured here: https://www.audiosciencereview.com/...box-x5-phono-preamp-review.59982/post-2209103. So it's lucky to clear a 70dB SINAD while Spartan 20 clears maybe 78dB or so. Oof. The real question is, just how good is the Spartan 20 if you drop an OPA1612 into it! You ought to try and resubmit it! Or just socket the opamp and put the proper one in the box.

 

[ryanmh1]

The Stereophile review of the Luxman E-07 probably warrants its own thread (once it makes it to the website), but suffice to say it's a product that would at least keep up with anything else measured to date (with a cartridge connected). Plus, it's audio jewelry. And, it's available for about $4k direct from Japan, which is fairly "cheap" (once you see the thing and what's in it). SNR for MM was somewhere around 80dB and 84dB when bandwidth was restricted from 20Hz to 20kHz. MC dropped by 10dB or less if memory serves, and the thing uses SUTs. Buckets of loading options, a heroic power supply, huge overload margins, the whole bit. Rarely would I call an audio component that costs as much as this a reasonable deal, but this is. I suspect it would easy cost a few thousand to buy the parts to make one.

Anyhow, the real point for this thread... Supposedly these levels were achieved with a discrete JFET front end. That is surprising since AFAIK Atkinson also measures without a load, where JFETs will usually "suffer" from the voltage noise (in a sense). But they paralleled four of them, and obviously piled on the feedback. Stacking them up would reduce the noise by almost 3dB or so, and there you have it. Presumably, they used some other tricks as well to pull that off with JFETs. It's exactly what I was hoping for from the above Cambridge, but alas. It was not to be. This is the level of performance we'd love to see for under $500--it is entirely achievable--but so far has not been found. If you have the magazine, check out the review if you really want to see what is arguably one of the best commercial phono stages out there.