I'm building headphoneamp using LM6171. Power supply rails are +/- 15V (LM317/337 regulated) and I'm looking for good caps for power supply, for each opamp's bypass. Schematic is Jan Meiers HeadWize - Project: A DIY Headphone-Amplifier with Natural Crossfeed by Jan Meier with some "upgrades", like more stiffer power supply.
Sanyo OsCon 47µF/16V are my first choise, but do these need any ceramic/plastic caps in paraller? Are there any better alternatives for less money..? Is 15V too high supply voltage for 16V OsCon's?
Sanyo OsCon 47µF/16V are my first choise, but do these need any ceramic/plastic caps in paraller? Are there any better alternatives for less money..? Is 15V too high supply voltage for 16V OsCon's?
are you opposed to SMD? actually the oscons are not even so great by todays standards, but they have a good name. the nichicon polymers are MUCH cheaper for the 47uf/16 you have part number PCG1C470MCL1GS at digikey which come in a leaded part as wel
but i much prefer the panasonic special polymer (SP-Cap), which is lower ESR and a better part overall than either the sanyo, or the nichicon. a bit pricier, but not as much as the sanyo and actually this range has some of the lowest esr caps of any polymer at <5mOhms however they are not rated as high in voltage. the murata polymer are rated for 12.5v in the 47uf and are actually lower ESR than the panasonic in that voltage rating. its only the lower voltage higher capacitance panasonics that are so obscenely low
may i ask why you want such high rails? would not +/-7.5-10v suffice? that would open up your options
but i much prefer the panasonic special polymer (SP-Cap), which is lower ESR and a better part overall than either the sanyo, or the nichicon. a bit pricier, but not as much as the sanyo and actually this range has some of the lowest esr caps of any polymer at <5mOhms however they are not rated as high in voltage. the murata polymer are rated for 12.5v in the 47uf and are actually lower ESR than the panasonic in that voltage rating. its only the lower voltage higher capacitance panasonics that are so obscenely low
may i ask why you want such high rails? would not +/-7.5-10v suffice? that would open up your options
just how "optimum" do you want?
LM317/336 have different output bypass optimums, if you have dual sec xfmr that provide all 4 wires then use 2 bridges, and a LM317 on both with identical parts
then the LM317 likes some ESR in the large electro output bypass to control noise, output impedance peaking - oscon/polymer electrolytic are "too good" - you should add a series damping R
old ~10uF tantalum electros were the default rec because they had ~1 Ohm ESR, often 25-50 uF Al eletros were given as alternatives - but modern Caps of both types now come with way too low ESR
http://web.archive.org/web/20080829161956/http://www.national.com/appinfo/power/files/f10.pdf shows the issue can be complicated
http://www.tnt-audio.com/clinica/regulators_noise2_e.html shows ref bypass, without big cap on output is fine with LM317 - if your load is ony inches from the reg, same pcb, broad traces then just skip the high uF output bypass cap
"insane" measurement based thread: http://www.diyaudio.com/forums/power-supplies/143539-another-look-lm317-lm337-regulators.html
LM317/336 have different output bypass optimums, if you have dual sec xfmr that provide all 4 wires then use 2 bridges, and a LM317 on both with identical parts
then the LM317 likes some ESR in the large electro output bypass to control noise, output impedance peaking - oscon/polymer electrolytic are "too good" - you should add a series damping R
old ~10uF tantalum electros were the default rec because they had ~1 Ohm ESR, often 25-50 uF Al eletros were given as alternatives - but modern Caps of both types now come with way too low ESR
http://web.archive.org/web/20080829161956/http://www.national.com/appinfo/power/files/f10.pdf shows the issue can be complicated
http://www.tnt-audio.com/clinica/regulators_noise2_e.html shows ref bypass, without big cap on output is fine with LM317 - if your load is ony inches from the reg, same pcb, broad traces then just skip the high uF output bypass cap
"insane" measurement based thread: http://www.diyaudio.com/forums/power-supplies/143539-another-look-lm317-lm337-regulators.html
Last edited:
16V caps should be fine on a 15V regulated rail. The advantage of using a higher voltage rating is a lower ESR component. Some prefer 100V rated electrolytics.
I breadboarded my headphone amp and ended up with 2 x 100uf 16V oscons per rail for each op amp, 8 caps in total. It was interesting to try different caps in the circuit. Other caps I tried were Panasomic FC, Elna Silmic and some 100V 100uf caps.
I breadboarded my headphone amp and ended up with 2 x 100uf 16V oscons per rail for each op amp, 8 caps in total. It was interesting to try different caps in the circuit. Other caps I tried were Panasomic FC, Elna Silmic and some 100V 100uf caps.
You will have to be careful to not create any unwanted high-frequency resonances, by using caps with very low ESR, such as film, in the presence of nearby parasitic inductances (such as an electrolytic cap, and/or the supply rails).
For the small-value caps (if used), X7R ceramic is usually recommended instead of film. But if you insert some resistance, as I think GLooP suggested, then that possible-resonance issue might be alleviated somewhat.
The caps will primarily provide decoupling. i.e. Like a small point-of-load power supply, they will supply the transient current demands that might be impossible to get through the inductance of the supply rail, with accurate timing, and without causing an unacceptably-large voltage disturbance on the supply rail.
You will get better transient performance if you use multiple smaller capacitors in parallel, instead of one 47 uF capacitor, especially if their connections are also all paralleled all the way to where the decoupling is needed, i.e. not shared in common. That way, the inductances of the caps and the connections will also be reduced, in the same way that resistances reduce when placed in parallel, and you can get the inductance to be less than that of one capacitor and its connections, which only works if there is no mutual inductance. As a bonus, the total ESR will also be reduced in the same way.
You want the impedance, as seen by the supply/gnd pin pairs of the active device, to be low-enough, up to the frequency that corresponds to your fastest-possible rise time (which will most-likely be much faster than is required for a 22 kHz sine, by the way; maybe use the opamp datasheet "typical" value).
f = 1 / ((π)(trise))
If you choose a maximum allowable voltage rail disturbance value, dv, and figure out what the maximum transient current change, di, might be (probably the zero-to-rail voltage divided by the load resistance), i.e.:
di = rail voltage / load resistance (or use worst-case load impedance)
then you can calculate the target maximum impedance that should be seen by a power/gnd pin pair of the active device:
Zt = dv/di
If you use your now-known di, dv, and trise (dt), and an equation similar to the standard differential equation for a capacitor, you can calculate the capacitance needed in order to be able to supply a transient current that will slew by di amps in time dt while only disturbing the voltage rail by dv volts:
C = di (dt/dv)
I think that you can calculate the maximum allowable total inductance for that capacitance and its connections to the decoupled points by using an equation similar to the standard differential equation for an inductance:
Lmax = dv (dt/di)
You can probably estimate the actual inductance for each capacitor by summing the conductor lengths for its connections to the decoupled pins, and its lead spacing, and multiplying by around 15 nH per inch. If you get more than Lmax for the best-case layout and one cap, you might need to use parallel caps with parallel connections. But that's just the inductance. I think you'd still need to check whether or not you can meet the Zt target impedance requirement up to frequency f, too, which might be harder.
You would need to take into account the capacitances, parasitic inductances, and parasitic resistances, to see if you meet the target impedance up to frequency f. I usually just use the LTspice simulator for that, with AC Analysis with an AC=1 current source (where the active device would be) and a model of the decoupling caps and power supply rails and smoothing caps etc, including all parasitic resistances and inductances.
Cheers,
Tom
For the small-value caps (if used), X7R ceramic is usually recommended instead of film. But if you insert some resistance, as I think GLooP suggested, then that possible-resonance issue might be alleviated somewhat.
The caps will primarily provide decoupling. i.e. Like a small point-of-load power supply, they will supply the transient current demands that might be impossible to get through the inductance of the supply rail, with accurate timing, and without causing an unacceptably-large voltage disturbance on the supply rail.
You will get better transient performance if you use multiple smaller capacitors in parallel, instead of one 47 uF capacitor, especially if their connections are also all paralleled all the way to where the decoupling is needed, i.e. not shared in common. That way, the inductances of the caps and the connections will also be reduced, in the same way that resistances reduce when placed in parallel, and you can get the inductance to be less than that of one capacitor and its connections, which only works if there is no mutual inductance. As a bonus, the total ESR will also be reduced in the same way.
You want the impedance, as seen by the supply/gnd pin pairs of the active device, to be low-enough, up to the frequency that corresponds to your fastest-possible rise time (which will most-likely be much faster than is required for a 22 kHz sine, by the way; maybe use the opamp datasheet "typical" value).
f = 1 / ((π)(trise))
If you choose a maximum allowable voltage rail disturbance value, dv, and figure out what the maximum transient current change, di, might be (probably the zero-to-rail voltage divided by the load resistance), i.e.:
di = rail voltage / load resistance (or use worst-case load impedance)
then you can calculate the target maximum impedance that should be seen by a power/gnd pin pair of the active device:
Zt = dv/di
If you use your now-known di, dv, and trise (dt), and an equation similar to the standard differential equation for a capacitor, you can calculate the capacitance needed in order to be able to supply a transient current that will slew by di amps in time dt while only disturbing the voltage rail by dv volts:
C = di (dt/dv)
I think that you can calculate the maximum allowable total inductance for that capacitance and its connections to the decoupled points by using an equation similar to the standard differential equation for an inductance:
Lmax = dv (dt/di)
You can probably estimate the actual inductance for each capacitor by summing the conductor lengths for its connections to the decoupled pins, and its lead spacing, and multiplying by around 15 nH per inch. If you get more than Lmax for the best-case layout and one cap, you might need to use parallel caps with parallel connections. But that's just the inductance. I think you'd still need to check whether or not you can meet the Zt target impedance requirement up to frequency f, too, which might be harder.
You would need to take into account the capacitances, parasitic inductances, and parasitic resistances, to see if you meet the target impedance up to frequency f. I usually just use the LTspice simulator for that, with AC Analysis with an AC=1 current source (where the active device would be) and a model of the decoupling caps and power supply rails and smoothing caps etc, including all parasitic resistances and inductances.
Cheers,
Tom
Last edited:
If you care about subjective sound quality I wouldn't use Oscons and similar caps.
Such caps are great for digital decoupling, though.
LM6171/2 needs rail to rail decoupling, I suggest you a 10nF film/foil cap (not metallized), like Wima FKP2, FKS2, FKC2.
On rails to ground you can use an audiograde cap 47-100uF 35V or superior, I suggest Cerafines, avoid Nichicon KZ and Silmic II (Silmic I are good).
I'm suggesting 35V voltage rating since, usually, it's the break point to lower dissipation factor and ESR.
A lot of low-esr cap are quite good (like FC/FM) but compared to audiograde ones they always have something wrong (harshness, unbalanced, etc.)
Strictly IMHO.
and what do you mean by high impedance headphones? hd600 (300 ohms) will play well with about 4-5vrms with peaks of up to 7 enough for even the most staunch lover of dynamics. +/-9v is plenty unless using very old headphones higher than 600R, with an opamp that wont swing close to the rails. tantalum is not obsolete, but the worlds resources of it are running out, so its not as common and only found in fairly expensive caps these days. one of the caps i linked is a polymer tantalum
Not exactly on the topic, but quite close and very useful:
Analog Devices : Analog Dialogue : PCB Layout
Also, check out the AD797 datasheet, there is very useful application data there.
AD797 | Ultralow Distortion, Ultralow Noise Op Amp | Operational Amplifiers (Op Amps) | All Operational Amplifiers | Analog Devices
Analog Devices : Analog Dialogue : PCB Layout
Also, check out the AD797 datasheet, there is very useful application data there.
AD797 | Ultralow Distortion, Ultralow Noise Op Amp | Operational Amplifiers (Op Amps) | All Operational Amplifiers | Analog Devices
Ok, things are going more complicated. In my first prototype I tried AD797 and OP27. AD797 was better for specs but OP27 had better sound quality. I think better power supply will improve AD797 performance.
I'll make external power supply with 200VA 2x24V toroid and rectifier + caps and 24V pre-regulators (LM7824) and for amp PCB second regulators.
Another idea for PSU was using small lead-acid batteries (like 12V 4Ah) and 13,8V float charge.
I'll make external power supply with 200VA 2x24V toroid and rectifier + caps and 24V pre-regulators (LM7824) and for amp PCB second regulators.
Another idea for PSU was using small lead-acid batteries (like 12V 4Ah) and 13,8V float charge.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.