Here's a badger/blameless type amp, with two mods that each cut the simulated distortion by about 10db. Can you spot them?
Spectrum shows 9kHz + 10kHz tones, with in-band distortion floor a full 130db below signal peaks. That's into a 2 ohm load!!
ps. This circuit also works with slower drivers and outputs, like the 15032 / 21195 and complements. In that case distortion is around -120db.
Spectrum shows 9kHz + 10kHz tones, with in-band distortion floor a full 130db below signal peaks. That's into a 2 ohm load!!
ps. This circuit also works with slower drivers and outputs, like the 15032 / 21195 and complements. In that case distortion is around -120db.
Attachments
Last edited:
[SPOILERS FOLLOW :-]
The first mod is the EF 2.5 output stage. This is a regular EF2 stage plus an extra buffer between the VAS and OPS. In the schematic this buffer is Qef25, R5, and R14. Like a real EF3, this increases the impedance of the OPS observed by the VAS from ~50kOhm to ~500kOhm. The VAS transistor is still loaded by its own Early effect of ~100kOhm, but at least that doesn't have a lumpy zero-crossing region.
Like a real EF3, this boosts VAS transimpedance, providing more local feedback through the miller cap or TMC network.
Unlike an EF3, this is easy to add to an existing EF2 amp. It's only 3 elements -- an outpatient surgery.
Subjectively, the EF2.5 changes the character of a badger to be less mellow and more detailed. Differences are most noticeable with low-impedance speakers.
I originally posted about the EF 2.5 here.
The second mod is the dual-gain global feedback loop. The gain is about 20 in-band, and then C8 and R15 cause the gain to rise to about 200 at HF. This allows the use of 10x lighter compensation while retaining the same gain-crossover frequency. This allows for faster slew rate and more global feedback at treble frequencies.
Don't C8 and R15 add their own phase lag to the global feedback loop? Well, yeah. This is worst in the transition from low- to high-gain, and then the excess phase approaches zero at higher frequencies as C8 becomes invisible.
Having such high gain in the 1MHz region seems risky -- it could make the amp vulnerable to RF at the inputs. We can counteract that with a 2nd-order low-pass filter at the amp input. With the steeper filter in place, neither 1MHz gain nor the unity-gain frequency (about 2.5MHz) are changed much from the original badger/blameless design. Happily, the extra low-pass filtering also keeps the in-band response flat to 20kHz. Without it, in-band response would start rising before 20kHz by almost 1db.
I've built each of these mods in separate amps. I haven't yet combined them in one circuit so YMMV.
The first mod is the EF 2.5 output stage. This is a regular EF2 stage plus an extra buffer between the VAS and OPS. In the schematic this buffer is Qef25, R5, and R14. Like a real EF3, this increases the impedance of the OPS observed by the VAS from ~50kOhm to ~500kOhm. The VAS transistor is still loaded by its own Early effect of ~100kOhm, but at least that doesn't have a lumpy zero-crossing region.
Like a real EF3, this boosts VAS transimpedance, providing more local feedback through the miller cap or TMC network.
Unlike an EF3, this is easy to add to an existing EF2 amp. It's only 3 elements -- an outpatient surgery.
Subjectively, the EF2.5 changes the character of a badger to be less mellow and more detailed. Differences are most noticeable with low-impedance speakers.
I originally posted about the EF 2.5 here.
The second mod is the dual-gain global feedback loop. The gain is about 20 in-band, and then C8 and R15 cause the gain to rise to about 200 at HF. This allows the use of 10x lighter compensation while retaining the same gain-crossover frequency. This allows for faster slew rate and more global feedback at treble frequencies.
Don't C8 and R15 add their own phase lag to the global feedback loop? Well, yeah. This is worst in the transition from low- to high-gain, and then the excess phase approaches zero at higher frequencies as C8 becomes invisible.
Having such high gain in the 1MHz region seems risky -- it could make the amp vulnerable to RF at the inputs. We can counteract that with a 2nd-order low-pass filter at the amp input. With the steeper filter in place, neither 1MHz gain nor the unity-gain frequency (about 2.5MHz) are changed much from the original badger/blameless design. Happily, the extra low-pass filtering also keeps the in-band response flat to 20kHz. Without it, in-band response would start rising before 20kHz by almost 1db.
I've built each of these mods in separate amps. I haven't yet combined them in one circuit so YMMV.
Last edited:
Hi Shaq888,
I am interested.
Thank you!
@jpc2001,
Can you post a better image of your FFT plot? Perhaps an .asc file to see in simulation.
Thanks!
Sa muli,
Albert
Haven't found how to hide it under a spoiler tag.
Measurement conditions are next:
+24dBVrms(16Vrms) for 90Hz, 1kHz and 10kHz
+18dBVrms(8Vrms) for each tone in mix of 19.5+20.5kHz
Load is resistive, 3.3Ohm
All levels on spectra are absolute, so the main tones are located as above.
A few words about the schematic. Essentially it is based on Selfs blameless, but includes Sam Groners idea using mosfet in VAS buffering + some combination of two pole/transitional miller comp + a bit of lag-lead compensation.
Attachments
Here are the .asc for the schematic in the OP, another .asc which applies the dual-gain mod to a Rotel RB-970BX, and the corresponding .png for the Rotel.
I built the dual-gain mod into a real Rotel RB-970BX and it sounds a lot cleaner. I never loved the stock sound of that amp.
On the Rotel, R28 and C12 modify the gain at HF. Compensation is now an 82p cap at C611, and there's extra low-pass filtering at the input to prevent RF at input from causing trouble.
Compensation was formerly a 330p cap at C611 plus two 220p caps, one from each predriver's base to its supply rail, for a grand total of 770pF hanging off the VAS output node which must charge and discharge with the full output voltage swing. I removed both 220p caps so the compensation is almost 10x lighter now.
I built the dual-gain mod into a real Rotel RB-970BX and it sounds a lot cleaner. I never loved the stock sound of that amp.
On the Rotel, R28 and C12 modify the gain at HF. Compensation is now an 82p cap at C611, and there's extra low-pass filtering at the input to prevent RF at input from causing trouble.
Compensation was formerly a 330p cap at C611 plus two 220p caps, one from each predriver's base to its supply rail, for a grand total of 770pF hanging off the VAS output node which must charge and discharge with the full output voltage swing. I removed both 220p caps so the compensation is almost 10x lighter now.
Attachments
Another way to understand this is to consider that C8 & R15 provide positive feedback.
Certainly needs to done with care, some Tian Return Ratio plots would be instructive.
Best wishes
David
Post1 has two attachments. The left one is a schematic which is upside down compared to conventional postings.
Out of curiosity I simulated the amp.asc file. I replaced the capacitive test in the schematic with 8r load instead. The simulated plot is at 1KHz and roughly 57w output power.
I noticed rail voltage is 55vdc. Are the 2n5089 LTP miirrors up for the job?
Regards,
Albert
Attachments
Here you go.
I made some minor edits here, which don't change the basic concepts or the performance of this amp:
- replaced the JFET current source (which represented a current regulating diode, though that wasn't clear) with a more typical BJT current source
- more typical bias spreader
- beefier 15032/3 drivers are better for two pairs of outputs
Attachments
Good call, thank you. Stability analysis was necessary...
I've never done a Tian analysis, which looks complicated? Let's start by breaking the feedback loop at a low-impedance point, dusting off the old 1GH inductor and dropping that in.
Whoops: phase margin falls to only 35 degrees at around 200kHz, and then recovers at higher freqs. The worst-case phase margin happens here at 200kHz, not at gain-crossover frequency.
Phase margin must stay at 45 degrees or better, so this design must change. My next post will fix it.
Attachments
Setting HF gain to about 4x the in-band gain keeps the phase margin at safe levels. The worst point on the curve is at about 100kHz where phase margin falls to 49 degrees before recovering.
The loop-gain plot showed an opportunity to tighten the TMC values a bit, so I did. The result still simulates with -129db distortion into 2 ohms.
The loop-gain plot showed an opportunity to tighten the TMC values a bit, so I did. The result still simulates with -129db distortion into 2 ohms.
Attachments
Tian is not complicated to do if you just follow the instructions, most of the work has been done in the implementation, thanks to Frank Wiedmann.
Your posted loop plots look inconsistent with your claimed distortion, one or other is incorrect, or both.
Since the "enormous inductor" technique is known to be incorrect that would be the obvious place to start.
Use Tian, post your results and we can try to make sense of them, my first post was probably also not sensible.
Best wishes
David
Last edited:
BTW, in cases when your feedback divider impedance to output impedance ratio is big, you can use a simplified method, evaluating only a voltage part (if you insert the probe in the right place). It will give a minor error compare to proper Tian or Middlebrook method, but most likely it will be smaller than other inaccuracies in such case. That have to be used with caution, because sometimes it seems it is almost OK but it is really not, but it is almost always fine with audio power amplifiers.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.