Discrete Headphone Amp.

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I'm teaming up with a colleague at work for a discrete headphone amp, based on the circuit shown, targeting a Christmas present for my colleague's father. The design is Class A and employs a "fake SIT" in the output stage. Simulations are promising, and the first proto board is biasing up with only a 2-3mV offset at the output. I'll write more on this as the project proceeds. Packaging is TBD, but it'll need to be something fairly classy.
 

Attachments

  • SHR_HP_SCH.GIF
    SHR_HP_SCH.GIF
    27.8 KB · Views: 2,757
Last edited:
Member
Joined 2011
Paid Member
Seems like you'd want to make R17 adjustable, in order to set the desired bias current in output transistor Q9.

As drawn, the output bias current is a function of Q8_Vthreshold, Q9_Vthreshold, and Q1_Vthreshold. Since the datasheet spread (max - min) for Vthreshold is pretty large, the possible variation in Q9 bias current is pretty large too.
 
Pots were used to set the bias, then replaced with the proper fixed resistor. Two pots were used - one in place of R17 to set bias current, and one replacing R27 to adjust for minimum offset. Once this was done, actual offset was around 1mV.
 
Last edited:
I may look into another option of this design using Darlingtons instead of mosfets in strategic positions, which would make the bias easier to set up. I'll run a simulation to see if that approach is worthwhile. The current source for the input pair could be replaced with a BSS159, which has a tight spread of Vgs vs. Id. For this initial take, I used what I had on hand.

The input devices will remain mosfets or jfets.
 
Something screwy is going on here, as the simulation shows proper behavior, with appropriate amplitude overdamped square wave response and no funny business. More bias current gives a cleaner response. Food for thought and grounds for further research...

As a guide, the simulation waveform with the junk on its trailing edge is at 140mA bias current, while the clear waveform is at 240mA bias.
 

Attachments

  • SHR_SQW_140ma.GIF
    SHR_SQW_140ma.GIF
    47.9 KB · Views: 272
  • SHR_SQW_240ma.GIF
    SHR_SQW_240ma.GIF
    49.7 KB · Views: 249
Member
Joined 2011
Paid Member
I hypothesize there will be interesting data if you plot the Vds of PMOS source follower Q11 during an output rising edge event. I hypothesize that Q11 quite possibly might depart the "pentode" region (Vds>(Vgs-Vth)) and enter the "triode" region (Vds < (Vgs-Vth)) during a large amplitude, high slew rate, rising output edge.
 
I hypothesize that I won't be doing that, as I consider the slewing behavior of this circuit so far as pathological - frustrating, as it's not displayed by the simulation. The problem has been fixed in the real world, but it's my partner Shrut's turn to chime in re this situation. Soon all will be revealed, including some much cleaner slewing behavior (sigh of relief).
 
The time domain response Shrut shows in the last post drove me absolutely crazy for a while, as the downstroke has significant delay and sluggish response that was not evidenced in the simulations of this circuit ( input is the upper waveform, output is the lower trace). Since it's the downstroke that is the problem, I guessed it had something to do with the "buffered fake SIT" (Q10 and Q11) in the output stage. In the simulations, it made no difference whether the drain of Q11 was connected on the upstream or downstream side of voltage dropper Q12. On a hunch, I reconnected it to the downstream side as shown in the attachment, figuring that insufficient voltage across Q11 might give rise to the problems we were seeing with real-life square wave excitation. I leave it to partner-in-crime Shrut to reveal the results of this seemingly minor change....
 

Attachments

  • Shr_hp_sch.gif
    Shr_hp_sch.gif
    28 KB · Views: 613
Last edited:
Member
Joined 2011
Paid Member
Congratulations! Very nice pure exponential rising and falling outputs with no overshoot. Since the input is only 110 millivolts the amplifier is not driven into slew rate limiting, which means (a) you can't measure slew rate on this photo; and also (b) you CAN estimate the amplifier's bandwidth, from the well known bandwidth-risetime equivalency formula

0.34 = risetime_in_sec * bandwidth_in_Hz​

link to Wikipedia page

I'm seeing a rise time of about 5 microseconds (just an eyeball guess from the extremely zoomed-out scope photo @ 50usec/div) which implies a bandwidth of 68 kilohertz for your headphone amp circuit.

Most of the >50 watt power amps in Bob Cordell's book have a bandwidth of 500 kilohertz, FYI (also a gain of 21X). An NE5532 opamp, operated at a gain of 10.5X, has a bandwidth of 300 kHz. I looked up its bandwidth at gain=10.5X because that's gain of the schematic attached to post #17.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.