Yes, that's generally a bad idea but he's not doing that (according to the schematics he showed earlier).
Yes, that will happen when someone who knows something about electronics gets too close to someone who thinks bronze heatsinks sound better.
How about checking for DC before hooking up the speakers next time?
Last edited:
For newbies and those in North America, we tend to describe filters using frequency rather than time constant. The relationship is simply a reciprocal:
frequency = 1 / time
So:
frequency = 1 / 90 milliseconds = 11 Hz
frequency = 1 / 130 milliseconds = 7.7 Hz
So Andrew's saying to set your input RC filter to 11 Hz and the negative feedback filter (NFB RC) to 7.7 Hz.
As God correctly states 1/[2pi * RC time constant]
and 2pi ~= 6.283
giving F-3dB of the passive input filter ~ 1.6Hz (for my personal taste in what I think sounds right).
The amp must be lower than the passive filter and the filtering effect of the smoothing caps lower still.
This is where a bottleneck most often occurs.
The Smoothing cap RC >= sqrt(2)*NFB RC
and
The NFB RC >= sqrt(2)*Input RC
That puts the smoothing caps > +-20mF per 8ohms channel.
Skimp on the PSU filter and you should increase the frequency of the Passive input filter to match
Last edited:
Sigh ... thanks for the correction.
As if to prove my point--which I hope was not missed. When talking to electronics newbies (esp. N.A.) avoid/explain time constants. Input filter (Input RC) corner frequency = 1 / (2 * pi * 90 milliseconds) = 1.77 Hz
Feedback (NFB RC) corner frequency = 1 / (2 * pi * 130 milliseconds) = 1.22 Hz
(I thought 11 and 7 seemed high)
Andrew:
The Smoothing cap RC >= sqrt(2)*NFB RC
and
The NFB RC >= sqrt(2)*Input RC
Andrew is succinctly giving you the "big picture" here. Before we unpack the dense math (math's beauty) recall one of Andrew's key design philosophies:
Don't give a circuit a signal that it can't handle!
No frequencies that are too high or too low, here we're focusing on too low.
So your input filter sets a high pass filter at 90ms.
1.8 Hz (90ms)
The next circuit (NFB RC) has to handle 1.8 Hz well so set it to at least 1.414 [square root of 2, or sqrt(2) ] times 90ms. 127 ms (Andrew's 130 ms).
127ms is 1.25 Hz.
So the negative feedback circuit is set to easily handle the 1.8 Hz passed on to it by the input filter.
Now here's Andrew's really cool point! The 'big picture', the amp as an interconnected system, and connected to other systems (source, speakers, ...) ...
Can the actual amp handle 1.25 Hz well? The amp's ability to produce low frequencies interacts with the load (speaker, 8 ohm) to form a ... high pass filter!
This amp-speaker 'filter' should be set to at least 1.414 times 127 ms.
= 180 ms (0.88 Hz)
Amazing and elegant but ... what does it mean? Well, to have a filter you need a active element (cap or inductor) and for the amp this ultimately comes down to the big power supply caps (smoothing cap)! I bet you never saw your ps caps as such a simple filter before!
Now working backwards with our formula:
corner frequency = 1 / (2 * pi * time constant)
time constant = R * C
so
f= 1 / (2 * pi * R * C)
rearrange for C
C= 1 / (2 * pi * R * f)
C= 1 / (2 * pi * 8 ohm * 0.88 Hz)
= 22,600 uF (=20 mF)
On both the positive rail and the negative rail so Andrew's: smoothing caps > +-20mF per 8ohms channel
Now you know how much ps capacitance you theoretically need instead of just 'bigger is better'. A more level-appropriate explanation will sink in more!
Cheers,
Jeff
PS Careful with Andrew's cool easy formula. It is for time constants not freq! Then use
f = 1 / (2 * pi * time constant)
to convert to freq.
PSS (Oh, please let my math be right or I'll get blasted!
I got 2 hrs sleep last night ... for the 7th night in a row ...)
Last edited:
I and many who do the arithmetic find working with familiar RC time constants far easier than working with frequencies and having to convert to component values, repeatedly.
If you know that the Low Pass input filter is 0.7us then you can almost instantly work out the unknown component value from the known one.
Assume the Rs 1k ohms.
The filtering cap to attenuate the RF will be 0.7nF to give that 0.7us RC
680pF is close enough.
If you have Rs=200ohms and you insert an extra 1800ohms in series the total Rs forming the RF filter is 2kohms. The required cap is 350pF for that 0.7us. Use 330pF.
If you prefer 1us or 0.3us, the calculation for the component values is just as easy.
Convert your required F-3dB frequency to an RC time constant ONCE. Then use RC for all further work with that filter.
BTW,
Audio's explanation is really good, better than I can manage.
If you know that the Low Pass input filter is 0.7us then you can almost instantly work out the unknown component value from the known one.
Assume the Rs 1k ohms.
The filtering cap to attenuate the RF will be 0.7nF to give that 0.7us RC
680pF is close enough.
If you have Rs=200ohms and you insert an extra 1800ohms in series the total Rs forming the RF filter is 2kohms. The required cap is 350pF for that 0.7us. Use 330pF.
If you prefer 1us or 0.3us, the calculation for the component values is just as easy.
Convert your required F-3dB frequency to an RC time constant ONCE. Then use RC for all further work with that filter.
BTW,
Audio's explanation is really good, better than I can manage.
Yes, that's generally a bad idea but he's not doing that (according to the schematics he showed earlier).
Yes, that will happen when someone who knows something about electronics gets too close to someone who thinks bronze heatsinks sound better.
How about checking for DC before hooking up the speakers next time?
Oh, I checked and tested everything other than DC offset before hooking my first pair of test speakers up. But my issue didn't happen the first time I turned it on, it happened more like the 20th time I turned it on. After reading info AndrewT posted, I changed some things with the power supply, tested everything including DC offset, and again the amp didn't send anything other than sweet sounds to the speakers through many on-off clycles before the horrible hum shot through attempting to foil my listening efforts for a second time.
If I cannot get to the point where I know I don't have to check my amp for DC every time I want to turn it on I will scrap any idea I had around living with a Gainclone. Is that possible or is this Gainclone stuff a never ending nightmare of potentially blown speakers?
glenv6, just unplug any running chipamps at this time.
What you describe is strange, we should trouble shoot it. I think a separate thread with a title like "Chipamp: strange, intermittent DC offset!!11!" would be appropriate and help others with similar issues too (that's more that just 'bloated' !! )
It could be something strange like grounding, or turn-on order, quick on/off off/on, ... We need info: schematics, pictures, measurements ... we'll get to the bottom of it.
Oh, and no, chipamps are awesome!!! I properly designed chipamp is very safe and rugged. Remember, chipamps are in TVs, small stereos, gaming systems, small Home Theater systems, .... because they are robust and cheap ... sorry: cheap and robust!
I figured there was more going on once you mentioned the second blown speakers and didn't assume the worse like some posters.
...
Getting back to the OP:
after comparing my GC with quality commercial amps I've confirmed that my build sounds "bloated". By that I mean that there is more than enough bass and mid-bass, but mids and highs seem recessed. The amp obviously has more dynamics compared to others, lacks in HF transparency, but it sounds "bloated"
Build a Andrew approved chipamp and see if it sounds unbalanced.
Cheers,
Jeff
PS
Andrew: "BTW, Audio's explanation is really good, better than I can manage." That made my day!
What you describe is strange, we should trouble shoot it. I think a separate thread with a title like "Chipamp: strange, intermittent DC offset!!11!" would be appropriate and help others with similar issues too (that's more that just 'bloated' !!
) It could be something strange like grounding, or turn-on order, quick on/off off/on, ... We need info: schematics, pictures, measurements ... we'll get to the bottom of it.
Oh, and no, chipamps are awesome!!! I properly designed chipamp is very safe and rugged. Remember, chipamps are in TVs, small stereos, gaming systems, small Home Theater systems, .... because they are robust and cheap ... sorry: cheap and robust!
I figured there was more going on once you mentioned the second blown speakers and didn't assume the worse like some posters.
...
Getting back to the OP:
after comparing my GC with quality commercial amps I've confirmed that my build sounds "bloated". By that I mean that there is more than enough bass and mid-bass, but mids and highs seem recessed. The amp obviously has more dynamics compared to others, lacks in HF transparency, but it sounds "bloated"
Build a Andrew approved chipamp and see if it sounds unbalanced.
Cheers,
Jeff
PS
Andrew: "BTW, Audio's explanation is really good, better than I can manage." That made my day!
What you describe is strange, we should trouble shoot it. I think a separate thread with a title like "Chipamp: strange, intermittent DC offset!!11!" would be appropriate and help others with similar issues too (that's more that just 'bloated' !!
AudioLapDance - Good advice! I'll start a separate thread... Thanks!
-Glen
Boy, that escalated quickly Thanks guys!
Now I did some calculating based on all this, but first a few questions:
1) Should I use the stock R1 (1k) if I'm using an input cap? I presume I should, so I should calculate Input RC with 1k+22k = 23k. Right?
2) If R1 (Rs?) is 1k, then RF decoupling cap is 680pF - solder it at the input RCA or at the LM3886 pins?
Now for Input RC to be 1,7Hz, C should be around 4uF when R is a given 23k (22+1).
Using 3,3uF which I have at hand, the frequency goes up to 2,09Hz, and using that stock Ci of 47uF together with 680R the NFB RC would be 4,98Hz - but it should be 1,45Hz max!
Long story short - how does this sound:
Input RC: C=3,3uF, R1=1k, R2=20k (lower) >>> 2,29Hz (69ms)
NFB RC: C=100uF (higher), R3=1k (higher) >>> 1,59Hz (100ms)
69ms*sqrt(2)=97,6ms
The next problem are the PSU caps, which are 10mF per channel. I do have another pair of 10mF, but they are Jamicons general usage, so I'm not sure if I should use them or buy another pair of Panasonic TS...
PS Andrew - took me a while to see that the HF and MF decoupling caps are already in my amp, what you suggested is to solder them directly to the pins, not add new ones... Will try to do that, though the MF caps legs are probably too short now.
Thanks guys!Now I did some calculating based on all this, but first a few questions:
1) Should I use the stock R1 (1k) if I'm using an input cap? I presume I should, so I should calculate Input RC with 1k+22k = 23k. Right?
2) If R1 (Rs?) is 1k, then RF decoupling cap is 680pF - solder it at the input RCA or at the LM3886 pins?
Now for Input RC to be 1,7Hz, C should be around 4uF when R is a given 23k (22+1).
Using 3,3uF which I have at hand, the frequency goes up to 2,09Hz, and using that stock Ci of 47uF together with 680R the NFB RC would be 4,98Hz - but it should be 1,45Hz max!
Long story short - how does this sound:
Input RC: C=3,3uF, R1=1k, R2=20k (lower) >>> 2,29Hz (69ms)
NFB RC: C=100uF (higher), R3=1k (higher) >>> 1,59Hz (100ms)
69ms*sqrt(2)=97,6ms
The next problem are the PSU caps, which are 10mF per channel. I do have another pair of 10mF, but they are Jamicons general usage, so I'm not sure if I should use them or buy another pair of Panasonic TS...
PS Andrew - took me a while to see that the HF and MF decoupling caps are already in my amp, what you suggested is to solder them directly to the pins, not add new ones... Will try to do that, though the MF caps legs are probably too short now.
the HF decoupling must be attached with a very short round trip route = short leads and short traces.
The MF decoupling can have a longer route. The Pi filter effect benefits from the extra resistance and extra inductance of the longer route. But the round trip route still needs to be reasonable. 50mm to 100mm of total round trip route would be acceptable. Whereas the HF total would be 5mm to 15mm.
The MF decoupling can have a longer route. The Pi filter effect benefits from the extra resistance and extra inductance of the longer route. But the round trip route still needs to be reasonable. 50mm to 100mm of total round trip route would be acceptable. Whereas the HF total would be 5mm to 15mm.
seems good. It needs smoothing RC>=140ms.
Last edited:
RF and interference attenuation
RF and interference attenuation:
I put the main input filter on the input to the amp PCB. There it can attenuate RF and interference coming from inside the Chassis and any external fields (entering via the cable Signal/Hot) that leak through the Chassis.
I also fit a VHF attenuator at the input socket to catch the very strong external fields as soon as they arrive at the equipment. This is a differential mode filter.
Recently I posted an enquiry about attempting to catch both the differential mode and common mode interference at the XLR input socket. But I have had no response to my suggestion and question.
RF and interference attenuation:
I put the main input filter on the input to the amp PCB. There it can attenuate RF and interference coming from inside the Chassis and any external fields (entering via the cable Signal/Hot) that leak through the Chassis.
I also fit a VHF attenuator at the input socket to catch the very strong external fields as soon as they arrive at the equipment. This is a differential mode filter.
Recently I posted an enquiry about attempting to catch both the differential mode and common mode interference at the XLR input socket. But I have had no response to my suggestion and question.
He posted this schematic in post 3, which shows a 220pF between +IN and -IN. That cap should go from +IN to A_GND, not -IN if you want a RF filter.
So this is the plan - desolder the 100nf bypass caps from the board, solder them together and directly at the V+ and V- pins as you stated before.
The MF decoupling caps are about 5cm apart on the board as is, so do you think it would be ok to leave them there? I should measure to be sure, but total route seems to be around 10cm. Another option is to desolder the cap leads connected to the ground and connect them to the HF ground with a trace as you stated before.
I'm just thinking how to solve this without buying new caps because of the short leads on the existing ones...
The smoothing caps are a problem, I will leave them at 10mf per channel for now until everything else is in place, listen to the amp, and then try with the extra 10mf Jamicons.
Hmm, I'm pretty sure I saw a schematic somewhere that had that cap in the same place. Is this a mistake then, is my 220pf cap doing any good or bad there?
There are a few articles/papers stating that great care must be exercised when attaching any capacitance to the -IN pin of an opamp. This applies equally well to chipamps. They are opamps with a heavy duty output stage.
The warning is there because some added capacitance on the -IN pin makes for worse stability.
There are some papers that show how to use added -IN capacitance on the -IN pin to improve stability, but the values are usually very low and usually not much bigger than the parasitic capacitance of a PCB.
The warning is there because some added capacitance on the -IN pin makes for worse stability.
There are some papers that show how to use added -IN capacitance on the -IN pin to improve stability, but the values are usually very low and usually not much bigger than the parasitic capacitance of a PCB.
And Andrew, what do you make of this reverse math, using the existing 10mf per channel and some standard values:
Smoothing caps: 10mf, Rspk=8R >>> 1,98Hz (79,5ms)
NFB RC: Ci=100uf, R3=560R >>> 2,84Hz (55,6ms)
Input RC: Cin=3.3uf, R1=1k, R2=10k (original is 22k) >>> 4,38Hz (36,1ms)
Is -3dB point of 4,38Hz too high?
And does changing R2 from 22k to 10k, and R3 from 680R to 560R affect the chip much? Although the changes are not big, so I guess it would work fine
Smoothing caps: 10mf, Rspk=8R >>> 1,98Hz (79,5ms)
NFB RC: Ci=100uf, R3=560R >>> 2,84Hz (55,6ms)
Input RC: Cin=3.3uf, R1=1k, R2=10k (original is 22k) >>> 4,38Hz (36,1ms)
Is -3dB point of 4,38Hz too high?
And does changing R2 from 22k to 10k, and R3 from 680R to 560R affect the chip much? Although the changes are not big, so I guess it would work fine
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.