J-Mo Mk II headphone amplifier

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.

rjm

Member
Joined 2004
Paid Member
Status: concept stage.

Prior art: this and this and tangentially to this.

Current activity: doing it up in ltspice, confirming operating points, circuit values, components, etc.
 

Attachments

  • jmo2 00.png
    jmo2 00.png
    19 KB · Views: 4,846

rjm

Member
Joined 2004
Paid Member
LTSpice simulation progressing well. To design using Spice is not my thing, but for optimizing the operating points and debugging creations, it's pretty great.

This circuit topology is terribly inefficient, you have to bleed for output power. What really adds insult to injury here though is the 4 V turn on voltage of the IRF510. This is the main reason that a 12 V voltage rail only gets you 1.4 V max output (<4 V p-p).

I could get around this by using a 2SK213. Maybe, I haven't checked recently to see if my source still have them. The thing is, all the nice Toshiba audio MOSFETs are obsolete. It's all switching applications now, and most of those are SMT. The IRF510, with it's low input capacitance, still seems like the best option when availability of factored in.

(perhaps also Vishay Siliconix IRLIZ14G)
 
Last edited:

rjm

Member
Joined 2004
Paid Member
A digression from LT Spice to Excel for a moment.

It's not too hard to plot up a quick and dirty model for the circuit behavior, and the results give a nice visual indication of what's going on.

The output power of this circuit is typically determined by the source voltage, the source resistance, and the load resistance (headphone impedance). For the rest it's just a matter of setting the voltage rail (V+) high enough to avoid clipping before the maximum output current is reached.

So then it comes down to what output power into what headphone load is desired, and what level of heat dissipation will be tolerated.

As an aside, this circuit will work very nicely with high impedance headphones, but will not work very nicely with both high and low impedance headphones. The circuit values can be optimized for one or the other, but sadly not both.

I am concentrating here on low impedance phones, 16 ohms especially as these seem very common these days.

The design center values (i.e. what I'm hoping to obtain in the real circuit) are >100 mW output into 16 and 32 ohm headphones and about 2-3 watts of heat. I'd also like to avoid voltage clipping even for 300 ohm headphones, the maximum possible output power is obtained there, too.

The worksheet is pointing to Rs 33 ohms, Vs 5.5 V, V+ 20 V, and a standing power dissipation of 2.8 W per channel.

The V+ is higher than I originally planned, but the heat is within budget. I will refine the circuit gain to match these new operating points and feed it back into LT Spice.
 

Attachments

  • plot1.png
    plot1.png
    67.9 KB · Views: 4,508
  • plot2.png
    plot2.png
    35.8 KB · Views: 3,567
  • plot3.png
    plot3.png
    61.2 KB · Views: 3,353

rjm

Member
Joined 2004
Paid Member
LT spice agrees nicely with the worksheet estimates.

Working through it, I realized that it might be possible to get rid of the voltage divider on the input: if the gain is increased a little, to about 12 dB, the natural source resistor bias of the JEFT puts the MOSFET gate at the right voltage.

That puts you at the mercy of the jfet to set the operating points however.
 

Attachments

  • jmo2 10b.png
    jmo2 10b.png
    18.8 KB · Views: 3,404
  • jmo2 10c.png
    jmo2 10c.png
    18.7 KB · Views: 1,077

rjm

Member
Joined 2004
Paid Member
Been using LTSpice to get a better handle on how the front end of the circuit works.

Given a JFET-PNP pair, I can configure the desired gain. There is a lot of inter-dependency though, as a result of the feedback and the complication that here the AC gain sets the DC operating points. (!) There doesn't seem to be a need for an input bias, as long as the gain needed to correctly set the operating point also happens to be one you can live with for the audio signal gain, too. Fortunately this is largely true, as long as the JFET model can be chosen freely to match the desired signal gain.

So I need to settle what JFET would be best to use, and, of lesser importance, which transistor. Or to put it another way, how much current do I really need to send down the input stage, and how should I split the currents between JFET and PNP devices?

By the way, I rather like this JFET-PNP compound device. The gain is highly configurable between 1-10 (100?), and the bandwidth in excess of 1 MHz. I could imagine using it in a phono stage, with a second JFET replacing the output MOSFET.

Ideas ideas...
 

Attachments

  • J-Mo2 10h.zip
    44.1 KB · Views: 148
Last edited:

rjm

Member
Joined 2004
Paid Member
Up and running under simulation.

Gain 5.5x (15 dB), bandwidth ~1 Mhz, pushing just over 4 V p-p into a 16 ohm load, or over 11 V p-p into 300 ohms.

This is just before the onset of hard current clipping, the negative waveform is already heavily distorted (2nd harmonic) in both cases.

R3 adjusts the current through the JFET and hence the bias voltage across R4 which in turn fixes the MOSFET operating point. It can be made a 10k trim pot to allow full adjustment and channel matching.

P.S. Everything is ticking over quite smoothly. My only concern at this point that for low circuit gains of 3x or 4x the JFET gate-source voltage has to be fairly high, about 2 V, to lift the MOSFET gate up to the 9.5 V required. This generally means operating the JFET at an atypically low I_ds.

P.P.S. PSRR is 40 dB at 120 Hz, still 34 dB at 20 kHz. Not bad, considering it's a simple single supply circuit. Some sort of RC or LC filter stage or voltage regulation will be required. V+ ripple will need to be under 100 mV to avoid audible hum.
 

Attachments

  • J-Mo2 11a.zip
    971 bytes · Views: 124
  • Jmo2 11a.png
    Jmo2 11a.png
    50.5 KB · Views: 1,059
  • Jmo2 11a ii.png
    Jmo2 11a ii.png
    47.1 KB · Views: 676
Last edited:

rjm

Member
Joined 2004
Paid Member
This circuit isn't quite "build-and-go", the circuit values are worked out as follows. The main issue is that multiple resistance values have to be optimized for a particular JFET. A near substitution of JFET model can probably be cleared just by trimming R3, but there is a risk of inadvertently choking off the current to the pnp transistor if R3 is set too small without changing R4 and R6 to compensate.

1. Choose the output stage operating bias and V+, defined by Vs and Rs, depending on the desired output power, headphone load, and thermal budget.

-> Vs 5.5 V, Rs 30 ohms, and ~~ 3W for V+ 20V.

2. Mosfet gate is Vs + 4 V = 9.5 V. The input stage gain x the JFET source bias voltage must equal 9.5 V.

-> JFET source bias can be from 0-Vturnoff. Choose a highish current JFET like the J111 with a high Vturnoff, so that he gain doesn't have to be to high to bring the output up to 9.5 V.

3. The J111 (and similar with Vturnoff 2.5 V - 3 V) seem to run about 1 mA at a source bias of about 1.8 V - 2 V. The current through the pnp transistor Q2 should be 2 mA or higher or more to make sure there it has enough control to keep everything working as designed.

-> This means that R3 is about X, and, to keep the gain and transistor currents in the right ballpark, R4 should also about X and R6 is about 5X. First roughtly estimate X based on the JPEG used, for the J111 it's about 680-950 ohms. After setting R4 and R6 from this estimate, the final tuning of the operating points can be done by trimming R3.

The J111 seems like a widely available model, so we'll run with that. Any n-channel JFET with Vturnoff of around 3 V should be able to be substituted. Using a smaller Vturnoff JFET is possible but R4,R6 will have to be adjusted to increase the gain since the JFET bias is not going to go much above 1 V.
 
Last edited:

rjm

Member
Joined 2004
Paid Member
Matching JFETs

So, short story is this circuit is going to need well-matched pairs of JFETs. Which means buying 10 and picking the best 2.

Attached is a visualization for what to expect when buying 2N5484-5-6 series of parts. These are all the same JFET, just binned by the manufacturer depending on how they test.

Green circles indicate roughly the "most likely the part will measure somewhere in here" area.

Now, for the desired gain of 5-6x, and MOSFET gate of 9.5 V, the Vgs operating point should be about 1.5-1.8 V, and we are hoping for about 1 mA, so the target specifications are a Vd of 3-3.5 V and Idss of 5 mA. This is unliklely to be found for either the 2N5484 or 2N5485 parts. So either I compromise or look for another JFET.

Right now I'm thinking the 2N5485 looks pretty good, chances good of finding a couple around Vd = 2.5 V and Idss = 8 mA. That should be workable.
 

Attachments

  • jfet.png
    jfet.png
    35.2 KB · Views: 691
What I always like about the original was the unity gain without feedback, made a nice transparent substitute for an expensive OPT (hybrid tube amp.)

Always want to try and run complimentary with split supply and possibly eliminate the output cap, increase the input impedance so the input cap could be smaller less load on the tube.

Thanks for the update and the model.
 

rjm

Member
Joined 2004
Paid Member
Everyone wants to get rid of the output coupling cap. I tried. At the end of the day I don't think its worth the effort - if you want direct coupled output, a complementary buffer stage using bipolars is simply the easier way to go. Been there, done that. With the single-ended MOSFET buffer, I'm prepared to live with the coupling cap as part of the baggage that comes with using this particular topology.

As a result of the parameter spread as well as the thermal drift, FETs are a headache. Bias networks and AC coupling caps are the Aspirin. Any circuit that relies on these FET parameters to set something reliably is asking for trouble. That's one of the reasons why I'm being extra careful with the paper-build before I even bother to order some parts and play around.

Latest LTspice results appended. This is the version optimized around the 4V Vp 2N5486. It's a relatively chunky FET and all the currents are increased accordingly. As far as the distortion calculations go, however, the results are identical with the 2N5485 optimized circuit: what I considered the point just before the onset of hard clipping, output 3dB (2V pp) into 16 ohms, gives -26 dB 2nd harmonic (~~5% THD). That's 120 mW or so output power. The harmonic progression is textbook simple, as expected for a single ended circuit.

Once slighty away from hard clipping, the distortion spectrum cleans up significantly with a falling progression of 2nd, 3rd and 4th harmonics, each peak -25 dB below the last.

At 1/10th the output voltage (1 mW output power) the 2nd harmonic is under -73 dB, so even though the simulation doesn't compute it, the 3rd harmonic is about -100 dB and everything else is in forget-about-it territory.

Simulations like this are pretty useless except to show problems. The conclusion from the above is that it's basically sound.
 

Attachments

  • ltspice 120218.png
    ltspice 120218.png
    49.2 KB · Views: 1,055

rjm

Member
Joined 2004
Paid Member
A fairer, and more correct assessment of the performance away from clipping is the 2nd harmonic is always about -50 dB lower than the fundamental, so the THD is 0.3 %, pretty much independent of output power. 90% of that is second harmonic. 3rd order and higher harmonics total about 0.03%.

The 0.3% 2nd harmonic is relatively high and probably colors the sound. This is most likely a characteristic of the MOSFET + source resistor single-ended output topology, and explains the mellow, distinctive sound of headphone amplifiers based on it.
 

rjm

Member
Joined 2004
Paid Member
Updated ltspice file attached.

Spent some time figuring out how much tolerance I have in the JFET parameters, and which model JFETs can be used.

The restriction is I want a particular gain, I want the trim resistor R3 to be within range, and I want about 5 mA total current running through the input circuit (i.e. R4).

The other point is the JFET has to be a current, widely available model.

In the end I chose the J112. The J202 is a perfectly good substitute, and, in a nod to the original circuit, a 2N4416 can be also be used also if one takes into account the old school TO-18 package. All those should yield a decent proportion of useable devices.

The circuit shows a test jig that can be used when matching and testing JFETs. Bias the FET with a 2.2k source resistor and the source should autobias up to a voltage that is some statistical scatter. Anything in the range of 1.9-2.5 V (optimal is 2.1 V) should work in the main circuit just relying on R3 to make the final adjustement. The closer the JFETs can be matched between channels, the better.

The final word for now is: this is a nice update to the original, adding full 12 dB of gain and optimizing (as best as possible) for the full range of headphone impedances at the cost of only one additional transistor and - this is important - no loss of performance. The noise and distortion of both circuits are determined almost entirely by the MOSFET stage, and, just like the Szekeres they are based on, are not great. It's intrinsic to way this circuit works that there is a lot of 2nd harmonic distortion. These headphone amps are about achieving a nice, pleasant tone rather than the last word in transparency.
 

Attachments

  • J-Mo2 11h.zip
    1.4 KB · Views: 112
  • j-mo2 11h.png
    j-mo2 11h.png
    48.4 KB · Views: 1,016
Last edited:

rjm

Member
Joined 2004
Paid Member
Eagle .sch file attached. No .brd file yet, you are one your own there...

Schematic has some extra resistors added, some are for safety and some are for stability, some like R1 and R6 are probably not needed, but don't hurt.

R11OPT provides an optional set of pads when using a TO-247 package resistor for R11. Replace as needed if you want to use those Vishay RTO-20 power resistors or similar.
 

Attachments

  • j-mo2 11d.png
    j-mo2 11d.png
    21.9 KB · Views: 1,033
  • J-Mo2 11d.zip
    79.8 KB · Views: 130

rjm

Member
Joined 2004
Paid Member
Update on JFET parameters.

I'm matching devices by simply putting 2k2 on the source, grounding the gate, and applying 10V or so on the drain. The source auto-biases up to a voltage V, and I'm looking for sets around V= 2 V, implying a working current of 1 mA.

So far the J202's I bought run a little lean, V= 1.3 V, while the 2N5486's almost perfect at V = 2.5 V. The scatter on the 2N5486s is fairly broad however, but it's not too hard to get pairs within 100 mV.
 

rjm

Member
Joined 2004
Paid Member
Some thought on the power supply.

Circuit V+ calls for 20 V and 400 mA with a maximum ripple of about 10 mV.

(The PSRR is only about 40 dB, which along with a maximum allowable output ripple of 0.1 mV dictates that the V+ ripple should be no more than 10 mV.)

A capacitance input filter is impractical. ~~150,000 uF required. A CLC or LC filter is doable, but you are looking at a 1kg chunk of Hammond iron. (c.f. 159Y 600 mH 11 ohm).

The simplest route is CRC, C=4,700uF, R=10 ohms. The voltage drop is about 1-2 V, so the power transformer secondary winding should be about 15-16 VAC. I would probably use an 80 VA toroid.

A voltage regulator or capacitance multiplier is also possible, but it's worth keeping in mind that each volt dropped across the regulator dissipates almost 200 mW per channel. Vin = 30V, Vout=20V is 4W of heat for a stereo circuit. Modest heatsinking will be needed. (in the order of 10 C/W or better).

Options run from a 16V, 20V or 24V switching supply (common and cheap, sound horrible) to the LM317T (boring! sound is average) to a Z-reg, to a capacitance multiplier.

I would like to use my standard power supply modules (160VA 2x12VAC). This, in series, gives 28 V rectified. (I live in Japan, 100V line, so the output is 15% less than you'd expect since my transformers are 2x115 primary).

I'm leaning towards a MOSFET based Z-reg. Just because the gate voltage is conveniently 24V, i.e. 2x 12V zeners, which I have already. And the IRF510 I already have because it's part of the JMo2 circuit.

This amp, because of the low PSRR, lives and dies by it's power supply.

The fun part, though, is it's easy to swap different regulator/filter stages in and out, and see what works best.
 

rjm

Member
Joined 2004
Paid Member
Note on JFET matching

J202, 10V, Rs=2.2k, 25 samples

Bias current was found to range from 0.507 mA to 0.582 mA. The average is 0.55 mA, std. dev. is 0.02 mA.

From 25 samples, it is possible to obtain 7 pairs matched within 0.5%, or 11 pairs matched within 1%. Clear sailing.

The distribution is a truncated shape, clearly the parts have been "binned". Lower and higher current parts have been removed to end up as J201 or J203, respectively. The J202 parts tend to cluster towards the high current end, which helps when it comes to finding closely matched pairs.
 

Attachments

  • J202.png
    J202.png
    11.6 KB · Views: 857
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.