Simulation of the JE-990 OP Amp By Deane Jensen

The JE-990 (990) OP amp by the late Deane Jensen is an impressive piece of engineering. I was highly impressed when I first read the AES article:

and the application note for the LM-394:

I still like the JE-990 to this day.

Jensen is known for their audio transformers and some background about the company and founder is given here:

The 990 was first described in 1980 as a public domain design in the AES Journal:
"JE-990 Discrete Operational Amplifier"
JAES Volume 28 Number 1/2 pp. 26-34; January/February 1980

It was designed for use with Jensen transformers, and has found it's way into many Pro-Audio applications. It is a fairly standard OP amp topology, diff pair input loaded by a current mirror, standard VAS with a Beta multiplier, Baker clamp, pole-zero Cdom compensation, and additional HF compensation on the emitter resistor. The LM-394 Supermatch Pair transistor is used in the diff amp which provides low DC offset, low noise and distortion. It is essentially a low power amp that will easily drive 75 ohm loads.

The LM-394, current mirror load, Beta multiplier, Baker clamp, and pole-zero compensation were all innovative for the time.

John Hardy is probably the best known supplier of the JE-990, and also produces several pro-audio mic preamps based on Jensen transformers and the JE-990. He has some nice documentation with the JE-990 schematic here:

Edit: Why is it that only the first post can be edited? I'd like to fix the broken links here.
Last edited:
John Hardy posted some background on the 990 and Jensen Twin Servo mic preamp here on 21st November 2005, 09:46 AM :

And repeated here:

Some general comments about the 990, Deane Jensen, Steve Hogan and myself. The original JE-990 design was done by Deane Jensen and was finished in 1979. The engineering paper that Deane wrote about the JE-990 in 1979 was turned into an AES Preprint that appeared in the AES Journal in 1980. You can go to and pay for this preprint and download it, or I can send a copy to you if you provide a postal mailing address. In the latter half of 1979 I developed packaging for the 990 that was compatible with the API-2520 and offered it as a product. Various mic preamp cards followed over the next few years, with the M-1 Mic Preamp coming out in 1987.

The 990A and the 990C were developed in the mid/later 1980s, certainly in time to be used in the Jensen Twin Servo 990 Mic Preamp that I began manufacturing for Deane toward the end of 1988 (or so). This version of the Jensen Twin Servo Mic Preamp was a combined development of Deane, Steve Hogan, Bill Whitlock and myself, using the M-1 package as the foundation. It replaced the Twin Servo that was built for Deane from 1986 to 1988 by the Boulder Company.

Steve Hogan was instrumental in the development of the 990A and 990C versions. He has been in touch with Beno May about Beno's modifications, but I have no knowledge about the details of those modifications.

The 990 (or any op-amp) does not have "30dB of gain", or even "28dB of gain". It has whatever amount of gain you design it to have. There are limits to how much gain an op-amp can provide, and they are determined by a variety of factors, including open loop gain, the desired bandwidth, maximum allowable distortion, etc.

I never had a specific discussion with Deane about the reason for the two stage design of the Jensen Twin Servo. But I am almost certain that he looked at the combination of the JE-16-A (as it was known at the time) input transformer and the JE-990, and thought to himself something like this: "Wow, the JE-16-A is a perfect match for the noise characteristics of the JE-990, and it's my best input transformer because it has the lowest impedance ratio (150:600 ohms) of all my input transformers, but the low ratio only provides 5.6dB of voltage gain compared to 20dB of gain that the JE-115K-E provides (150:15k-ohms). If I need 60dB of gain, a high-ratio transformer would provide 20dB, the op-amp following it would provide 40dB. But the JE-16-A only provides 5.6dB, so the 990 will have to provide 54.4dB to get me up to a total of 60dB. That is a lot of gain for one op-amp (relatively speaking). I think I'll add a 2nd 990 in series with the first one so that each 990 will provide 27.2dB of gain for a total of 54.4dB of gain from the TWO op-amps." So each 990 in the Jensen Twin Servo mic preamp is adjustable from about 6dB to 27.2dB, for an overall gain range of about 18-60dB (including the transformer voltage gain).

In the M-1 mic preamp, the 990 is adjustable from 5.6dB to 54.4dB of gain, for an overall range of about 12-60dB including the transformer voltage gain.

I have modified quite a few M-1 and M-2 mic preamps, and a few Jensen Twin Servo mic preamps, for greater than 60dB maximum gain (the standard "maximum" gain). If you look at the schematic for the M-1 mic preamp on page 7 of my M-1 data package:

You will see that R7 determines the maximum gain of the 990. It is listed as 20 ohms, but it has actually been 19.1 ohms for many years. One approach to increasing the maximum gain is simply to add another 19.1 ohm resistor on the bottom of the p.c. board in parallel with the original 19.1 ohm resistor. This cuts the resistance in half, providing 6dB additional gain. It is easy to add the resistor, and easy to remove it if desired. I have put two of those resistors on a few preamps to increase the gain by 10dB.

So, how much gain is too much gain? Hard to say. YMMV. But the Twin Servo design might have a performance edge when venturing beyond 60dB because the gain is being split between two op-amps, each one working at a much lower and more relaxed gain. On the other hand, it seems that customers that have M-1 preamps with the high gain modification are happy with the results.

Hopefully this clears some things up. Thank you.

John Hardy
The John Hardy Co.
I was bored one afternoon so I quickly simmed the JE-990 in LTspice, the project file is attached as a ZIP file for those who might be curious. Semis are correct except for LM-394, most models are from Fairchild, outputs from OnSemi, MAT02 from AD.
Used the MAT02 as a substitute for the LM-394, and I only included the transistors not the diodes.

The simulation looks good, input source is set to overdrive the amp at 20 kHz in order to view the overload performance.

Pete B.


    6.5 KB · Views: 658
mlloyd1 said:

thanks for sharing. do you have any similar info you could share on the later versions of the 990 modules (eg. 993)?


Sorry no, that's the Boulder module as I understand it? No, I've not seen the schematic or anything much about it.

There was this thread,

The post below from that thread mentions a few of the improvements made by Boulder. I agree that input bias compensation is a good feature, however there have been DC coupled mic preamps at least made with the stock JE-990. A FET input would probably make even more sense to eliminate input offset current, and optimize the noise performance for higher impedance loads. Higher slew rate is easy if we lift the requirement for unity gain stability since the Cdom cap can be made smaller. Cascoding the diff pair also makes a lot of sense to minimize the non-linear Miller capacitance seen at the input. Most of these would be on the top of my list for improvements:

purer said:

The Boulder 993 Gain Stage
The sonic signature of any audio component depends largely on what circuit is used for achieving gain and buffering outputs. From the very beginning, rather than seeking any particular characteristic sound, the goal of maintaining sonic neutrality has been held high by Boulder.
The 990 is a discretely built gain stage, developed in the Hollywood audio community in 1978 under the coordination of the late Deane Jensen. It became widely recognized as the best universal gain stage ever, and was quickly retrofitted into many existing recording consoles at great expense.
Designed around a revolutionary matched dual transistor fabrication process, the 990 set a new standard for low noise and distortion.
Because the original 990 could not be DC coupled, Boulder added bias injection when it started using the 990 design and called it the BA990DC.
DC coupling eliminates the need for interstage capacitors. Otherwise, these form multiple high-pass filters and become a source of bass clarity problems.
Widely used for two decades, the 990 has indeed earned its reputation for critical music recording and reproduction. With the coming of the digital age, additional improvements make it the best choice for another decade or two.
In the new version, called the 993, Boulder doubled the slew rate, thus yielding the lowest distortion numbers in the industry. A special cascoded front-end circuit increases input impedance, eliminating this potential source of distortion.
All this is done without losing the phase margin and stability that the 990 is famous for. (Stability is absolutely essential for sonic clarity.)
The Boulder 993 is housed in a new metal case that provides extra shielding. DC performance is further improved by a special thermally conductive encapsulating compound.
The 993¡¯s capabilities include high slew rate, wide bandwidth, low distortion, high output current, and low output impedance. This combination cannot be matched by any integrated circuit.
The 993 continues to be the basis for new Boulder Series 2000 products into the next millennium.
The JE-990 uses inductors across the diff pair emitter degeneration resistors. This is to reduce LF noise and to increase the DC gain, and it is patented #4,287,479.

This patent should not hold up in court seeing as there was prior art as can be seen on page 38 of the .pdf in this link. Very nice work by Walt Jung in that paper. It is the ADI 121 Op amp designed by Dick Burwen in 1966, note also that it has input offset current compensation:

Note also the ADI 44 with FET inputs, on page 40, but otherwise it looks as if the JE-990 borrowed the rest of the design. I never noticed this before.

Pete B.
Problem with MJE172/182 models

I tried a simulation run to measure the THD at 20 kHz with a 10 Vpp output into 600 ohms. The THD was .05 to .07 % which seems high for an advanced design like this. Note that I added a large (1000 uF) cap in the shunt path of the external feedback network for these sims to minimize DC offset at the output.

Checked the output bias, and the MJE172/MJE182 run around 50 uA quiescient current - this is not right. Tried MJE243/MJE253, MJ15034/15035, MJE340/350 and the quiescient current was 20 to 30 mA - much better, and the distortion was reduced by one to two orders of magnitude.

All the output models I tried are from OnSemi but it seems to me that the MJE172/MJE182 are defective. I checked the models and they are the current ones at the OnSemi site.

I then tried Fairchild output models for the BD139/140, MJE340/350 and these provided 15 to 20 mA.

The OnSemi BD139/140 provided 80 mA, which seems high.

The OnSemi MJE172/182 and BD139/140 do not seem right, but more work would have to be done to determine the problem.

This is just a preliminary look at this issue, just in case anyone is using the simulation. There are obviously several work arounds, find better models for the correct devices, or I would suggest using the MJE243/253 (similar vintage) or MJ15034/35 (modern) models from OnSemi.

I should have a look at the JE-990 sim at the Yahoo LT-Spice users group, and see what they use for outputs when I have some free time.

Pete B.


2010-06-08 10:55 pm
PB2: Just download the file and opened in in LTSpice IV.
I wanted to see what it would show for input noise and got something in excess of 100nv/Sqrt(Hz) !
Wondering if you or anyone has revised your work to analyze noise? Maybe the modes used don't include noise? I used:
.noise V(Vout) V3 oct 10 20 20000
Hi, just noticed your post, sorry for the delay.
I don't recall if the models have noise parameter info or not.
Have you read the original article on the design, the noise versus source
impedance is covered, IIRC.
It would also be interesting to look at the spectrum of the noise.
I've not done anything with noise analysis for this design.
I can't seem to edit my previous post.

Add: .inc Cordell-Models.txt on the schematic sheet.

Then right click on the text for the old NPN output transistor and enter BD139C, same for PNP with BD140C. Took me a while to remember this since clicking on the part and using select new device from the menu of devices doesn't work.

The new output stage idle current is 14.75 mA which is a good amount of current, results in about .7W idle for the output stage with 24V supplies.

The MJE243/253 as outputs result in 16.4 mA idle current which is also reasonable.
Last edited: