Nelson Pass inspired headphone amp

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I just finished a headphone amp inspired by many of Nelson Pass's designs. It's basically a JFET son of Zen front end with a paralleled B1 buffer like the Beast with 1000 JFETs, but with only 5. It runs class A with global feedback and a total gain of 2. It's targeted for use with 32 ohm headphones which don't need alot of voltage gain, but can use the current. It's running at about 50ma per channel, so it dissipates about 2 to three watts. The JFETS do get a bit toasty - about 65 degrees C (155 degrees F). It does sound very nice however - I'm hearing details I've never heard before and now I think I'm held back by my source.

I used two 5v to 9v DC-DC converters to provide power. They operate at 1MHz, so it's well above audibility. The 5v is input through a mini USB port, so the power source can be a wall wart or computer. Electrolytic caps are 270uf OSCONs on the power supply and on the output. There are also 2.2uf film caps on the output and 4.7uF tants on the power supply. All JFETs are matched 2SK170.
 

Attachments

  • headphone.gif
    headphone.gif
    16.9 KB · Views: 3,440
  • hpamp.jpg
    hpamp.jpg
    70.1 KB · Views: 3,294
Interesting circuit - and great build job getting it all into the Altoid can!

Idea - maybe re-spinning it to use DN2540 depletion mosfets would help with the hot jfet issue. The N3 version (TO-92s) are rated at 1W and up to 120mA. The two jfet current sources would be replaced by depletion mosfet sources with a 100R gate resistor and somewhere around a 43R source resistor. The source resistor would set the current to 50mA, 5x the 10-12mA the 2SK170 jfet's datasheet curves seem to show at Vgs=0. So a single depletion mosfet would do what the 5 jfets in parallel are doing, saving a bit of board space. You could even go with the (TO-220) N5 version of the DN2540 for the two output FETs which would dissipate better. Those are rated at 15W and 500mA but otherwise have all the same specs as the N3 version. Mouser has the Supertex stuff.

http://www.supertex.com/pdf/datasheets/DN2540.pdf

An introduction to depletion-mode MOSFETs (current source in fig 7, minus the 100R gate resistor that should be there)

Just a thought. :)
 
Last edited:
Looks good.

Are your diodes the right way around though in your circuit?

They're LEDs and I may have drawn them the wrong way around in the circuit, I'm always forgetting which direction they go in schematics. They are the right way in the physical circuit. They're old 3mm low brightness LEDs, those seem to have the lowest AC impedance. Plus they provide a nice visual indicator that the power is on.

I thought about other devices for the CCS, but the JFETS just seemed the most logical. A single TO-92 device of any sort may have dissipated too much on its own - spreading it across 5 devices increases surface area and reliability. The heat dissipation will always be the same since it's running Class A. Putting some space between the devices might be nice for air movement, but 65 C is still a fairly good temperature by discrete component standards. Plus it seemed more in line with the whole "bunch of JFETS" theme.
 
AX tech editor
Joined 2002
Paid Member
Fenris,

That's a nice circuit! Having held the actual 1000- beast board, I have a weak spot for similar implementations!
I see that the gain of the circuit depends on the pot setting - whatever the equivalent pot impedance is, it's in series with the 10k feedback resistor.
That does modify the actual volume law when turning the pot.
What value of pot did you use?

jan didden
 
The heat dissipation will always be the same since it's running Class A. Putting some space between the devices might be nice for air movement, but 65 C is still a fairly good temperature by discrete component standards. Plus it seemed more in line with the whole "bunch of JFETS" theme.

You are right - that eventually hit me later on. :p Total dissipation would be the same. And probably no practical way to mount the TO-220 versions on the back of the board to heat sink out to the can.

Well congrats again on the build! :)
 
Fenris,

That's a nice circuit! Having held the actual 1000- beast board, I have a weak spot for similar implementations!
I see that the gain of the circuit depends on the pot setting - whatever the equivalent pot impedance is, it's in series with the 10k feedback resistor.
That does modify the actual volume law when turning the pot.
What value of pot did you use?

jan didden

It's an Alps linear 20K (3B curve), which is high enough for almost every source (CD player, DAC, sound card, iPod, etc). In practice, the overall volume control is fairly linear in the middle, although it does lose some sensitivity at the end of its travel. Higher values of feedback resistors would help, since that's the lowest value of pot in that model line.
 
Doing the math, the high value of the pot is really limiting my overall gain. I'm probably getting only about a third of the gain (and 1/10 the power) I could. My breadboard prototype used a 2K pot which had minimal impact on the feedback/gain relationship. I'm going to do a quick fix by changing the 1M input grounding resistor to 4.7k, that should give me about twice the voltage gain and four times the power output. Looking at the curves, this will greatly linearize the voltage output and render the power output a nice, shallow exponential increase. As it stands now, the voltage output levels off and the power output only increases linearly (which doesn't correspond to how our hearing works). If I were to do it over again I'd use 47K for the feedback resistor, 20K for the input series resistor, and a 4.7K for the grounding resistor
 
Member
Joined 2009
Paid Member
Doing the math, the high value of the pot is really limiting my overall gain. I'm probably getting only about a third of the gain (and 1/10 the power) I could. My breadboard prototype used a 2K pot which had minimal impact on the feedback/gain relationship. I'm going to do a quick fix by changing the 1M input grounding resistor to 4.7k, that should give me about twice the voltage gain and four times the power output. Looking at the curves, this will greatly linearize the voltage output and render the power output a nice, shallow exponential increase. As it stands now, the voltage output levels off and the power output only increases linearly (which doesn't correspond to how our hearing works). If I were to do it over again I'd use 47K for the feedback resistor, 20K for the input series resistor, and a 4.7K for the grounding resistor

With this mod: "47K for the feedback resistor, 20K for the input series resistor, and a 4.7K for the grounding resistor & 10K log pot" can drive 300 ohms loads like Sennheisers HD600-650? wich Idss values for matcthed SK170, are BL grade?
 
Last edited:
the 2SK170s are BL grade for the source follower (the 5 in parallel). I don't remember which type are in the gain stage, but it's low current anyway so it doesn't matter. Just make sure that the JFET acting as a constant current device is lower in IDSS than the other. For a 300 ohm load, you'll need more voltage gain and less current. A voltage gain of 6 would be about right. I'd go with a 330K resistor for the feedback and a 47K for the series input. I'd also run at least +/-12v rails and change the resistor in the LED string to about 1K. Ideally, the voltage taken at the top of the resistor should be about -50mv. The value can be calculated by knowing the IDSS of the constant current JFET. Take your negative power supply voltage, subtract the voltage drop across the LEDs, and size the resistor according to the current.

For 300 ohm headphones you will need more voltage and less current. But the power handling ability of the JFETS is limited. You may want to reduce the current through the JFETS acting as current sources for the source follower. This will keep them within their dissipation limits.
 
I'll actuall ammed what I said earlier. It might be better to just go with the standard BOZ style resistor biased front end similar to this:
http://www.diyaudio.com/forums/pass-labs/103050-jfet-boz-15.html#post1832422
and eliminate the feedback. Overcomplicating this circuit might not be best, even if it has less theoretical distortion.

I'm going to install RMAA and test things out in the next few days.
 
How much did it cost?

Would it be possible to power AKG K271 MKII with this?

Easily. The output impedance is below 10 ohms and depending on gain and load can output over 100mw. The K271s are more efficient than most and have a nominal impedace of around 55 ohms. I'd recommend a bit more gain (maybe about 4). I'm even thinking about a different version with a MOSFET and LM317 CCS as the source follower (to reduce parts count). I'm also looking at a DC servo to eliminate the output capacitors.

Cost isn't that high, about $15 for the JFETs, $6 for the DC-DC converters, $20 or so for the rest of the passives, and about $15 for the PCB. Cost could be cut $10-20 by going with MOSFETs and 317s and a DC servo (eliminating the bigger OSCONs). I'm still playing with this on my days off and am trying to get some decent measurements.
 
At the moment, this isn't for a novice. The resistors are all 0805 surface mount, which isn't the easiest to solder. You more or less need a double sided PCB made since everything is so compact. If you want to use perfboard and through-hole components this could be done but it would take up more space. It also depends on what kind of power supply you want to use - the DC-DC converters I used might be a bit expensive to ship to Sweden.

I've got to work for the next 7 days straight. After that I'll get some time to work up a perfboard layout and try to use commonly available components.
 
I decided to do a minimalist version of this schematic with the idea of a low cost, easy to source design for the first-timer. Here's the design I came up with. It has a JFET configured for a gain of 2. The design assumes an Idss of the JFET of approximately 10ma. It is designed to be run at approximately 60% of that in order to stay in the most linear region. That would be 6ma which would equate to 3.35v across the 560 ohm resistor. The adjustable pot that defines the virtual ground is adjusted so that the output at the virtual ground is .1v lower (this is the Vgs drop in the datasheet). The output of the JFET feeds a directly coupled MOSFET (IRF610) source follower with a 80ma constant current source. The output is capacitively coupled. I built a prototype this week using "bugsplat" construction in an old external hard drive case. It's powered by a laptop power supply. I've still got some white noise and humming issues to solve, but the power output is perfect for headphones (very, very loud at maximum with 32 ohm headphones, but power is limited to 0.2 watts).

I've got pictures I'll post in a little while. If anyone has any advice on how to get rid of the hum and static (which may be related to my construction technique or power supply) it would be greatly appreciated.
 

Attachments

  • headphone.gif
    headphone.gif
    15.6 KB · Views: 1,244
I've got pictures I'll post in a little while. If anyone has any advice on how to get rid of the hum and static (which may be related to my construction technique or power supply) it would be greatly appreciated.

I would guess it's the power supply being noisy and the amplifier having low PSRR. If so, put something between the supply and the amp to clean up the power rail. Something like CRC filtering or common mode choke on both V+ and Gnd.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.