Measured effect of op-amp rolling on a phono equalizer input-shorted noise

[GXAlan]

Op-amp rolling the stereo x LUXMAN LXV-OT10

Background
Last month, I put together a Luxman "kit" phono equalizer and posted the measurements.

Luxman LXV-OT10 Phono Equalizer Amplifier Review and Measurements

Luxman LXV-OT10 Phono Equalizer Amplifier Review/Measurements Let's take an audio company with nearly 100 years of history. That's Luxman, which has been around since 1925. Now take one of the oldest audiophile magazines in Japan. That's stereo, which was first published in 1963 -- just...

www.audiosciencereview.com

TLDR? The Luxman saturates quickly and early with its tube buffer, so it's not intended for transparency and adds a ton of harmonic distortion. It actually sounds great in moving magnet mode with my Sony PSX-555ES/Shure V15VMR and my Sharp RP-117 (3 mV output) but it was unusable in moving coil mode with my Monster Cable Sigma 2000 MC (0.2 mV output) due to excessive noise overwhelming the signal at any listening level. To be fair, the manual does warn about using low output MC cartridges of 0.2 mV or lower.

In the discussion, @AnalogSteph stated "With this circuit I would explicitly condone opamp rolling. It is absolutely hopeless otherwise."

Let's test this out! AnalogSteph had recommended the OPA1656.

This chip doesn't fit in the common DIP8 packaging, but thankfully @Audiophonics sells a pre-modified OPA1656 for 5 €. Not only was that affordable, I could trust that the op-amps would be legitimate. Shipping from France to the United States is pricey, so I picked up few extra OPA1656's and also got the most expensive single-chip OP-AMPs that Audiophonics had at 25€ and 33€, the NJR MUSES01 (J-FET) and MUSES02 (Bipolar). The instruction manual includes an article by stereo magazine who specifically recommends the NJR MUSES line for op-amp rolling.

Test Setup
The tube dominates SINAD and noise is the audible problem, so for these measurements, I shorted the inputs and then used my E1DA Cosmos ADC Grade A by @IVX and Multitone by @pkane to measure the noise as described in μV. The E1DA Cosmos had been calibrated using a 20,000 count true RMS meter in the mV and V range, and I am relying on the linearity of the E1DA to accurately represent μV. At the very least, the images can be compared.

The E1DA Cosmos ADC has DC offset which affects the way that Multitone "peak" value in the upper right. Maybe in a future revision, @pkane can have a checkbox for the E1DA, where his 64-bit math can be used to do a high pass filter at 2Hz so that DC offset is ignored.

The LXV-OT10 has two op-amps in stock configuration, the TI TL072CP and the JRC 4558D. Per the user manual, the TL072 can only be replaced with a FET input op-amp while the JRC 4558D can be replaced with a FET or bipolar.

I made the quick switch recommended by @AnalogSteph. What does 5€ do?

What a difference! I did a quick listening test and this was great! It had taken me from something completely unusable to pretty good.

What happens if I swap the JRC4558D for another OPA 1656 from Audiophonics? Not as big of a difference, but it's slightly better.

Does the 33€ MUSES02 do any better than the 5€ OPA1656? Not really. Maybe 0.3 dB better A-weighted noise, but they're pretty similar in this circuit. There may be differences with music, but at least for noise, there's not a big difference.

People seem to love MUSES op-amps though. I wonder what happens if I put both in the circuit.

That's really impressive!

The measurements were so good that I wondered if I had broken something since the legs of MUSES op-amps are slightly splayed out and I had to straighten them to fit the socket. Throwing the phono equalizer back into my setup, and WOW. The noise was clearly lower and everything sounded great. I guess the question is what a pair of MUSES01 would look like or the MUSES01 with the OPA1656 would look like.

Subjective Impressions
It's vinyl and I'm running it through a lot of tube distortion. It's not going to be transparent, so take my opinion with a grain of salt. At the listening position, the final set of op-amps (MUSES01/MUSES02) was so good that I heard no surface noise or background noise at my listening position. But the simple 5€ swap was good enough to minimize the noise to what I would normally associate with

Conclusions
Op-amp rolling doesn't seem to make a difference in well designed DACs or headphone amplifiers

At least with the LXV-OT10, op-amp rolling makes a big difference. The question is if this is due to the 52dB gain of a moving coil phono amp where small amounts of noise from an op-amp matters? Or was the LXV-OT10 kit designed so strictly to a price point that they went with cheap op-amps and went as far as recommending op-amp rolling in the manual?

It is also interesting to note that the guaranteed specifications of the MUSES01 and MUSES02 don't really stand out. So far, most of @amirm's op-amp rolling tests have been with "discrete op-amps" -- but we know from DACs that discrete DACs have failed to match the performance of the integrated circuits from ESS and AKM when it comes to noise performance. Even Marantz and their "musical mastering DAC" talk about their proprietary implementation allowing them to have the digital filter and noise shaping they prefer but that it comes at the trade-off of noise, and it's their "achievement" to get the sound they want with good noise characteristics.

 

[Sokel]

Graphs seem horrifying at first glance but the difference is 15db at the worst case and the level of the worst one is already at -90db (with the best been at -104db)
Nevertheless is impressive to know what can one see when has the way to measure.

There's too many variables though to extract something solid out of this,nothing guarantees the same improvement in any other circuit unless one has a way to measure it.

 

[GXAlan]

Graphs seem horrifying at first glance but the difference is 15db at the worst case and the level of the worst one is already at -90db (with the best been at -104db)
Nevertheless is impressive to know what can one see when has the way to measure.

There's too many variables though to extract something solid out of this,nothing guarantees the same improvement in any other circuit unless one has a way to measure it.

The is a very clear audible difference when actually used as a phono amp. Looking at residual noise is a good way to do this but what I will do is measure with a microphone in room and share the recordings.

Agree about no two circuits being the same. I just thought it was interesting that the op amps rolling worked well. Hopefully AnalogSteph who shared the original recommendation can give more details on the why.

 

[GXAlan]

Sample Recordings

 

[D!sco]

15dB is massive. Makes me wonder how little is actually necessary to get some of these products to decent shape.

 

[AnalogSteph]

I just thought it was interesting that the op amps rolling worked well. Hopefully AnalogSteph who shared the original recommendation can give more details on the why.

Here's input voltage noise density for JRC's TL072 counterpart, the NJM072:

That's at least 13 nV/√(Hz) broadband (and note that it still beats the TL072 spec of 18), before adding any passives. A decent MC prepre will generally be around 1 nV/sqrt(Hz), that's 20 dB lower! The really fancy kind with a bunch of parallel discretes in the input may reach <0.7 nV/sqrt(Hz) broadband. Using a "barefoot" '072 is basically an insult to the task at hand.

For reference, a 20 ohm resistor (a decent approximation for an MC cartridge) will produce a voltage noise density of 0.57 nV/sqrt(Hz) worth of thermal noise at 295 K. For 3 kOhms, it's 7.0 nV/sqrt(Hz), and for 10 kOhms, it's 12.8 nV/sqrt(Hz)... and given that MM cartridges can go up into the tens of kOhms towards the upper end of the audible spectrum, you can see why an '072 would be a more acceptable choice there. (This is also why FET inputs tend to be preferred for MM, they have almost no input current noise to speak of.)

OPA1656 has a lot lower broadband noise (ultimately reaching <3 nV/sqrt(Hz)) but being a MOS part, exhibits quite a bit of 1/f noise. So that's why things remained noisy on the bottom end while the rest improved drastically.

From your results, I rather have to suspect that MUSES01 may be substantially better than what the datasheet suggests. Well, either that, or you accidentally plugged them in the other way round (i.e. MUSES02 first, MUSES01 second). In which case NJM4580 / NJM8080 ought to make a good budget replacement. They all have relatively modest input bias current for a bipolar (100 nA typ), which may be low enough to not upset the circuit overly much (at least in MC operation).

BTW, using a linear voltage scale for the Y axis kind of stinks.

 

[computer-audiophile]

Just a side note. I didn't know that this kit from Luxman existed, but I find it very interesting.

The amplifier apparently follows a similar concept to a phono preamplifier with tubes that I built myself, namely continuous variable equalisation. This is a great idea if you have recordings that are not RIAA-cut. I actually had a lot of fun with this type of phono amplifier.

At that time, I orientated myself on a historical circuit model from ELAC, which I developed further.
That was the ELAC circuit (Excerpt from the 'Funkschau' magazine 1955)

 

[GXAlan]

15dB is massive. Makes me wonder how little is actually necessary to get some of these products to decent shape.

Yeah and even weirder is that manual says, “you should roll these op amps” for the best performance.

I cannot tell if this is some sort of master plan to convince people that there are “golden ears.” You swap the op amp and wow, what a difference. Now you trust the reviewers at Stereo Magazine and the engineers at Luxman to make things that sound great.

OR if this is really reflective of how tight margins are even on a kit. I think the kit is made in Japan.

I rather have to suspect that MUSES01 may be substantially better than what the datasheet suggests. Well, either that, or you accidentally plugged them in the other way round (i.e. MUSES02 first, MUSES01 second)

The Muses02, being bipolar, only works to replace the 4558 so definitely not a typo.

I do believe that to be the case. The subjective evaluation of the op amps talks about low noise of the MUSES op amps but the spec sheets aren’t all that great. Measurement wise, they do seem to perform well!

BTW, using a linear voltage scale for the Y axis kind of stinks.

Good point. @pkane can this be made as a toggle for the future version of Multitone, or like many things, is there already a setting that I can use to have a logarithmic Y-axis voltage scale?

 

[pkane]

Good point. @pkane can this be made as a toggle for the future version of Multitone, or like many things, is there already a setting that I can use to have a logarithmic Y-axis voltage scale?

Try using dBV for vertical scale.

 

[D!sco]

“you should roll these op amps”

Imagine a speaker company that recommended replacing the woofer. Seems about the same, crazy.

 

[GXAlan]

Imagine a speaker company that recommended replacing the woofer. Seems about the same, crazy.

I guess the one difference is that this is a kit. When it came out in Japan, they offered packages with op-amps.

Since this is “stereo magazine” in combination with Luxman working together, and apparently this line of products was inspired by the pandemic and having projects-at-home, stereo magazine talks about adding a ground wire from the chassis to the motherboard to reduce noise with MC designs along with a list of various op-amps AND the subjective sound difference offered by the golden ears.

The ground wire is easy enough to try…

 

[kemmler3D]

15dB is massive. Makes me wonder how little is actually necessary to get some of these products to decent shape.

A $5- $30 increase in parts cost will typically lead to a $25 - $300 increase in retail price for you, the user. So if @GXAlan 's speculation about designing to cost is correct, that could be why. A $120 kit raising the parts cost by even $5-10 could raise the price enough to seriously hurt sales. They had to cut corners somewhere...

 

[D!sco]

On the other hand, there’s the apple dongle applying the economy of scale to good engineering.

 

[kemmler3D]

On the other hand, there’s the apple dongle applying the economy of scale to good engineering.

Indeed. If you are willing to place an order for a million pieces, you can get really good deals on parts. Apple probably pays the least of any manufacturer for some of its off-the-shelf parts. For something like this kit where we're probably talking a few thousand units at most, they are probably paying the maximum price / MOQ for these things.

 

[dougi]

Noise maybe dependent on source impedance though, depending on the opamp. I did something similar on a similar phono amp (but two stage RIAA) . I used representative carts attached and measured the noise but calculated the S/N for representative input levels (5mV mm, 0.5mV mc) for two opamp types in the first gain stage. Difficult to get one type of opamp good for both mm and mc.

"What about noise? After reading the LT1028 datasheet, it made for interesting reading about noise vs source impedance. Hence I measured noise with a cartridge attached, AT VM540ML for mm and AT OC9XEN for mc. Yes, I could have used resistors but attaching a cart was easier and perhaps more representative of use. I also obtained some (expensive) LT1028s and tried them (after soldering in DIL sockets):

• mm, NE5534, SNR 75dB, 78dBA
• mm, LT1028, SNR 65 dB, 67 dBA
• mc, NE5534, SNR 58dB, 64 dBA
• mc, LT1028, SNR 64 dB, 71 dBA

This seems expected for the LT1028, better noise for a low impedance mc but worse for the higher impedance mm.

Given the specs state a mm SNR of 68dB for mm and 70dB for mc, this is perhaps A weighted and for the LT1028 as was in the schematic. Close to what I measured."

 

[LTig]

Noise maybe dependent on source impedance though, depending on the opamp. I did something similar on a similar phono amp (but two stage RIAA) . I used representative carts attached and measured the noise but calculated the S/N for representative input levels (5mV mm, 0.5mV mc) for two opamp types in the first gain stage. Difficult to get one type of opamp good for both mm and mc.

Actually it's impossible. Opamps with very low input voltage noise suffer from high input current noise, and vice versa. You have to sum both noise types for all resistors at both + and - inputs with source connected.

"What about noise? After reading the LT1028 datasheet, it made for interesting reading about noise vs source impedance. Hence I measured noise with a cartridge attached, AT VM540ML for mm and AT OC9XEN for mc. Yes, I could have used resistors but attaching a cart was easier and perhaps more representative of use. I also obtained some (expensive) LT1028s and tried them (after soldering in DIL sockets):

• mm, NE5534, SNR 75dB, 78dBA
• mm, LT1028, SNR 65 dB, 67 dBA
• mc, NE5534, SNR 58dB, 64 dBA
• mc, LT1028, SNR 64 dB, 71 dBA

This seems expected for the LT1028, better noise for a low impedance mc but worse for the higher impedance mm.

Yep. For MC go for AD797 or LT1028 [1] if you skip additional ultra low noise transistors in front of the opamp. For MM there is a wider choice but the NE5534A has the lowest price and is good enough.

[1] Douglas Self does not recommend the LT1028 due to its input bias cancellation circuit which may lead to more noise if the inputs see different resistances, but my DIY MC phono preamp was dead silent, same as with the AD797.

 

[AnalogSteph]

Actually it's impossible. Opamps with very low input voltage noise suffer from high input current noise, and vice versa.

Well, people back in the '80s did it by running discrete input stages with a bunch of low-noise JFETs in parallel, not uncommonly 3 or 4 pairs in high-end preamps. (Of course input capacitance nonlinearity would not be great, but at phono levels...) So it is doable even if the economy strikes me as a bit dubious. You can get by much more easily by adding a dedicated discrete MC prepre with inexpensive bipolars instead, as some brands kept on doing.