Is there generally ever any negative effects on amp circuit performance from increasing bias within reason?
Playing around with the XELF sims show that with a 300r load the distortion is lowest with the stock bias, 32r loads benefits hugely from increased bias, optimum 32r bias has about twice the distortion for 300r, even with much higher input levels for 300r.
With real headphone I suppose distortion will increase globally, I wonder if 300r HP would start to benefit from increase bias or a 300r bias switch would be a good idea.
Of course that isn't even considering bias effect on speed, IMD, output Z etc.
In reality lower bias also means less strain on PSU which will have benefits of it's own.
Playing around with the XELF sims show that with a 300r load the distortion is lowest with the stock bias, 32r loads benefits hugely from increased bias, optimum 32r bias has about twice the distortion for 300r, even with much higher input levels for 300r.
With real headphone I suppose distortion will increase globally, I wonder if 300r HP would start to benefit from increase bias or a 300r bias switch would be a good idea.
Of course that isn't even considering bias effect on speed, IMD, output Z etc.
In reality lower bias also means less strain on PSU which will have benefits of it's own.
If you run it with an opamp, the opamp will determine the distortion.
If you run the buffer on its own, then higher bias means lower distortion.
But it is diminishing return.
We have had 500mA bias on the XELF with sufficient heat sinking and power supply to suit.
You are into power amp numbers, when you only need a few mW audio output.
But by all means play around and see what pleases you most.
Those MOSFETs will take 1A bias without problems.
Patrick
If you run the buffer on its own, then higher bias means lower distortion.
But it is diminishing return.
We have had 500mA bias on the XELF with sufficient heat sinking and power supply to suit.
You are into power amp numbers, when you only need a few mW audio output.
But by all means play around and see what pleases you most.
Those MOSFETs will take 1A bias without problems.
Patrick
And here some crazy numbers :
Pro iCAN by iFi audio | Professional Studio Grade Headphone Amplifier
14W output power into 16R.
Power consumption 22W idle.
Patrick
Pro iCAN by iFi audio | Professional Studio Grade Headphone Amplifier
14W output power into 16R.
Power consumption 22W idle.
Patrick
The fact XELF can easily double up a low distortion speaker buffer is a really awesome bonus, gave me some motivation me to finish a speaker build.
Im starting to think that amp 'sound quality' (i.e distortion you actually hear vs the THD, IMD numbers) is defined mostly by open loop performance.
A good example is OPA1688 and OPA1622, both suitable for driving headphones with vanishingly low in distortion on paper. I've played around with these chips a lot in the past and to me the OPA1688 always sounded a lot worse compared to OPA1622.
OPA1622 is obviously designed primarily for HPs, one datasheet spec that stands out is the open loop output z, low and flat throughout audible range. OPA1688 has the same varying OL output z you see in most other audio op amps.
Im starting to think that amp 'sound quality' (i.e distortion you actually hear vs the THD, IMD numbers) is defined mostly by open loop performance.
A good example is OPA1688 and OPA1622, both suitable for driving headphones with vanishingly low in distortion on paper. I've played around with these chips a lot in the past and to me the OPA1688 always sounded a lot worse compared to OPA1622.
OPA1622 is obviously designed primarily for HPs, one datasheet spec that stands out is the open loop output z, low and flat throughout audible range. OPA1688 has the same varying OL output z you see in most other audio op amps.
To use the XELF for speakers you will need a dual opamp in DIP8 to lower the output impedance.
If you want unity gain, then use something like 2x OPA627 on an adaptor.
The best opamp is one that has an open loop bandwidth of > 1kHz.
There is no need for a lot of NFB. 40dB is plenty.
The OPA656 would have been ideal but for the low rail voltage.
The THS4631 is another good candidate.
But you should look around more.
Patrick
If you want unity gain, then use something like 2x OPA627 on an adaptor.
The best opamp is one that has an open loop bandwidth of > 1kHz.
There is no need for a lot of NFB. 40dB is plenty.
The OPA656 would have been ideal but for the low rail voltage.
The THS4631 is another good candidate.
But you should look around more.
Patrick
Here you can see what we did with the 1622 years ago.
New Audio Op Amp - OPA1622
New Audio Op Amp - OPA1622
New Audio Op Amp - OPA1622
New Audio Op Amp - OPA1622
Patrick
New Audio Op Amp - OPA1622
New Audio Op Amp - OPA1622
New Audio Op Amp - OPA1622
New Audio Op Amp - OPA1622
Patrick
I'm curious to try without the op amp first and see how it sounds.
Considering most tube amps it still wouldn't be that bad, especially for a ZFB amp.
Considering most tube amps it still wouldn't be that bad, especially for a ZFB amp.
Sorry for my cluelessness but when adjusting for speaker bias levels is there an important reason to adjust CCS current as described in post #6 or using Rbias alone is ok?
Really my plan is to use low voltage supply and high bias for HPs and an appropiate high voltage one (SMPS probably) for speakers which will also scale up bias current without any other adjustments (offset did not drift significantly in spice doing this, but i'd watch out for it in practice)
Really my plan is to use low voltage supply and high bias for HPs and an appropiate high voltage one (SMPS probably) for speakers which will also scale up bias current without any other adjustments (offset did not drift significantly in spice doing this, but i'd watch out for it in practice)
You have 2 channels on the board.
R_bias will allow you to adjust bias individually for each channel.
CCS current allows you to step increase / decrease bias on both channels simultaneously.
I would not use the CCS to do fine adjustments.
And you should not need it unless you want to switch between headphone amp and power amp modes all the time.
I did it just to demonstrate feasibility.
Also watch dissipation of the SMD devices.
I try to keep them below 30~40mW whenever I can.
Patrick
R_bias will allow you to adjust bias individually for each channel.
CCS current allows you to step increase / decrease bias on both channels simultaneously.
I would not use the CCS to do fine adjustments.
And you should not need it unless you want to switch between headphone amp and power amp modes all the time.
I did it just to demonstrate feasibility.
Also watch dissipation of the SMD devices.
I try to keep them below 30~40mW whenever I can.
Patrick
Sorry to bother you for the tenth time, the BOM mentions 220mA bias per channel but LTspice sims it as only 80mA (40 per rail).
The BOM says 2 for quantity of current regulating diode, maybe meaning their is 2 in series for 3mA total, but this would make bias too high according to sim.
The BOM says 2 for quantity of current regulating diode, maybe meaning their is 2 in series for 3mA total, but this would make bias too high according to sim.
> The BOM mentions 220mA bias per channel but LTspice sims it as only 80mA (40 per rail).
The Spice model in post #11 shows a bias of 0.6A as intended for power amp use.
When reducing Rbias (R9 in the spice model in post #11) to 1.7k, spice gives you exactly 200mA bias.
Of course this depends on the FETs you are using, hence the need to adjust.
> The BOM says 2 for quantity of current regulating diode
My mistake. You only need 1x 1.5mA.
Patrick
The Spice model in post #11 shows a bias of 0.6A as intended for power amp use.
When reducing Rbias (R9 in the spice model in post #11) to 1.7k, spice gives you exactly 200mA bias.
Of course this depends on the FETs you are using, hence the need to adjust.
> The BOM says 2 for quantity of current regulating diode
My mistake. You only need 1x 1.5mA.
Patrick
For Rbias a and b I was going off:
'510R nominal for Exicon 560R nominal for Renesas'
If these are not values for Rbias a and b what are they?
'510R nominal for Exicon 560R nominal for Renesas'
If these are not values for Rbias a and b what are they?
Vgs off of e.g. 2Sk1058 varies from 0.15V to 1.45V according to datasheet.
There is no universal value of R_bias that willl cover all possible MOSFET variations.
This is fact of life with FETs; their bias voltages vary from device to device, unlike BJT.
The MOSFET models used in the Spice file is just one particular Vgs value provided by Renesas.
It is not reality.
When you have the FETs and measured Vgs at bias, I can give you a closer value.
Patrick
There is no universal value of R_bias that willl cover all possible MOSFET variations.
This is fact of life with FETs; their bias voltages vary from device to device, unlike BJT.
The MOSFET models used in the Spice file is just one particular Vgs value provided by Renesas.
It is not reality.
When you have the FETs and measured Vgs at bias, I can give you a closer value.
Patrick
I have MOSFETs, actually they are selected EXICONs with yellow mark.
Would that give you an idea of approximate value?
Would that give you an idea of approximate value?
If you do not want to measure, then wire a 2k trimpot as variable resistor as R_bias and adjust till you get the bias you want.
Then measure the actual value and replace with fixed resistors.
Patrick
Then measure the actual value and replace with fixed resistors.
Patrick
R_bias in real life is calculated as follows :
I_CCS * (R_bias // TR1) = Vgs NMOS @ Bias + Vgs PMOS @ Bias + I_bias * R11
Just simple Ohm's law.
Patrickm
I_CCS * (R_bias // TR1) = Vgs NMOS @ Bias + Vgs PMOS @ Bias + I_bias * R11
Just simple Ohm's law.
Patrickm
- Home
- Amplifiers
- Headphone Systems
- The XEN XELF Headphone Buffer