ac coupled non-inverting amplifier
68k/2.2k or 68k/2.7k for gain setting
At NFB cap, 220uF//0.5uF, both electrolytic
Use whatever input load best suits the source device
At power, 100nF//330uF//2,200uF or larger, all low impedance caps
dc coupled non-inverting gainclone
27k/820R for gain setting
10k input load
4.7uF//10nF input filter cap
At power, 100nF//470uF//2,200uF or larger, all low impedance caps
Which is better? The ac coupled amplifier has more interesting dynamics, better speaker safety, and a longer life.
68k/2.2k or 68k/2.7k for gain setting
At NFB cap, 220uF//0.5uF, both electrolytic
Use whatever input load best suits the source device
At power, 100nF//330uF//2,200uF or larger, all low impedance caps
dc coupled non-inverting gainclone
27k/820R for gain setting
10k input load
4.7uF//10nF input filter cap
At power, 100nF//470uF//2,200uF or larger, all low impedance caps
Which is better? The ac coupled amplifier has more interesting dynamics, better speaker safety, and a longer life.
Yes, if you change some other component values accordingly.
22k/0,68k -> 33,35 times gain. 180k/10k -> 19 times gain. Which is better, apples or pears?
This is not DC coupled.
I have been calling it mixed AC & DC coupling.
True enough! Its only DC coupled when somebody forgets the input cap and roasts their speakers.
I'd have to regard DC coupled and Mixed coupled as "not interesting" because the AC coupled amplifier is more lively albeit a Very tiny bit more effort to construct.
Moving on to AC coupled. . .
After making several truly excellent AC coupled amplifiers, thanks to you, now I don't need any more DC coupled (or mixed coupled) amplifiers. Its just a personal choice--I'd rather have my dynamics and speakers intact, with a proper AC coupled amplifier. Sure, selecting an NFB cap for gentle clarity takes about 5 tries or so. But the payoff is fantastic.
68k? That's important for what it doesn't do. It is merely the averaged value of the practical low, 27k and the practical high, 114k, so the amount of current with 68k is correct, but I wasn't able to calculate more complex matters, such as phase.
Would you suggest a more optimized value than 68k and what would that be?
That would be the feedback resistor practical values when used with gain settings of 20~35 in a non-inverting LM1875 amplifier.
27k to 114k is a practical range for feedback resistor, from real amplifiers that are currently in service. A figure of 68k feedback resistor is merely a crude average.
As a bonus with 2.2k to 2.7k feedback-shunt resistor, an NFB cap is very easy to select and need not be larger than the amp.
In this example, what the 68k feedback resistor will do is [unlike 180k] promote a good size sound field, but [unlike 22k] there's not a hard sound artifact. That's certainly not the best of both. Its just directly in-between, with none of the problems.
You'd probably get the same/similar results with 47k, 56k, 75k, (feedback resistor) and its easily possible that one of those might be somewhat better than 68k. And that was my question--is a nearby value going to do better?
EDIT:
The original request was for a feedback resistor value better than 22k or 180k. Well, now that wasn't a tall order.
Last edited:
The NFB cap should be similar to the input resistor to ground to achieve low DC offset and good common mode rejection. For a non-inverting amplifier that leads to practical values between 10k and 100k. What you call the feedback-shunt resistor will then be sized according to the gain you want or need.
The sonic differences you describe could rather be the result of different gain settings, than of the absolute feedback resistor value. The higher the gain, the lower the bandwidth. The roll-off may already start in the audible range when the gain is high enough. Although I think you need bat ears to hear much of a difference. If you have those, you will be obliged to test whether nearby values give better results on your own. The difference will be below 0,1 dB which is usually considered too insignificant to derive particular sonic properties of one resistor versus the other.
They could also be a result of inadequate compensation that could lead to oscillation for a certain gain range. So you might need a capacitor in parallel to the feedback resistor for one gain, a different capacitor for another gain while a third gain setting may work without any capacitor. With the right capacitor values you should achieve the same sonic properties at least for a range of gains.
The sonic differences you describe could rather be the result of different gain settings, than of the absolute feedback resistor value. The higher the gain, the lower the bandwidth. The roll-off may already start in the audible range when the gain is high enough. Although I think you need bat ears to hear much of a difference. If you have those, you will be obliged to test whether nearby values give better results on your own. The difference will be below 0,1 dB which is usually considered too insignificant to derive particular sonic properties of one resistor versus the other.
They could also be a result of inadequate compensation that could lead to oscillation for a certain gain range. So you might need a capacitor in parallel to the feedback resistor for one gain, a different capacitor for another gain while a third gain setting may work without any capacitor. With the right capacitor values you should achieve the same sonic properties at least for a range of gains.
Depends on the PCB, the feedback resistor values and the IC you get. Compensation is also for the parasitic capacitance the board introduces. The capacitor size depends on the resistors, because it creates a filter together with them. And the typical gain-bandwidth-product may not be the one your IC actually has.
The LM1875's typical GBWP is given with 5,5 MHz, so a gain of 43 would (in theory) give a bandwidth of 127,9 kHz, while a gain of 31 would result in 177,4 kHz. You need a filter that reduces the bandwidth to 127,9 kHz to achieve the same roll-off. Assuming you set the gain of 31 with a 68k resistor, you use the well-beknownst formula C = 1 / (2*PI*f*R), where R is the feedback resistor. In theory you need an 18 pF cap, but since you are a good boy, you use an NFB blocking cap, and those two caps form a voltage divider which leads to a totally different frequency response than predicted, so you now need another cap between the two input legs which you know as Cc from the LM3875's or LM3886's datasheet to counteract that. To do the math from here on you need information about the amp IC and the PCB which you don't have, and will probably never get, and you will need measuring equipment and testing that is likely to end with a cap that is not 18 pF at all.
You could simulate the roll-off with a simple low-pass filter at the amplifier's input, but then again you do not know the gain-bandwidth-product of the IC you actually have. While the typical value given on page 2 of the datasheet is 5,5 MHz, the graph on page 4 indicates a value around 4 MHz and the limit is not specified as e.g. in the LM3886's datasheet where they say typical 8 MHz, limit 2 MHz.
Just keep in mind that the absolute value of a single resistor is not the decisive factor, but that the values of all components must fit to each other.
The LM1875's typical GBWP is given with 5,5 MHz, so a gain of 43 would (in theory) give a bandwidth of 127,9 kHz, while a gain of 31 would result in 177,4 kHz. You need a filter that reduces the bandwidth to 127,9 kHz to achieve the same roll-off. Assuming you set the gain of 31 with a 68k resistor, you use the well-beknownst formula C = 1 / (2*PI*f*R), where R is the feedback resistor. In theory you need an 18 pF cap, but since you are a good boy, you use an NFB blocking cap, and those two caps form a voltage divider which leads to a totally different frequency response than predicted, so you now need another cap between the two input legs which you know as Cc from the LM3875's or LM3886's datasheet to counteract that. To do the math from here on you need information about the amp IC and the PCB which you don't have, and will probably never get, and you will need measuring equipment and testing that is likely to end with a cap that is not 18 pF at all.
You could simulate the roll-off with a simple low-pass filter at the amplifier's input, but then again you do not know the gain-bandwidth-product of the IC you actually have. While the typical value given on page 2 of the datasheet is 5,5 MHz, the graph on page 4 indicates a value around 4 MHz and the limit is not specified as e.g. in the LM3886's datasheet where they say typical 8 MHz, limit 2 MHz.
Just keep in mind that the absolute value of a single resistor is not the decisive factor, but that the values of all components must fit to each other.
With Rod Elliot's headphone adapter, I can probably listen and guess from there and see about the 18pF as well as higher or lower.
While the ear cannot detect some sounds that are overall too quiet, if instead two given samples are loud enough to hear them, it can indeed detect very tiny harmonic differences as small as 0.03db. A temperature probe on the heatsink, gathering average temperature data, can tell the rest of the story, whereby average cooler is better stability. Clearer and cooler is the win.
PCB? There's only 5 pins for solo or 10 pins for parallel. No need to make it harder with a PCB. The 2-chip parallel amps are especially good because its a simple, fast, rigid assembly that cannot easily spin off from the insulator pad. Simply bend the v- almost to the chip(s) face(s) and then bend the v+ pins(s) half that distance. Set a 35v low impedance 470uF in from the left, repeat from the right. The ecap pins become the rails, horizontally across the chips. Add 100nF caps to stabilize the "free" ends, one on the left and one on the right. It is expediently fast and simple to do. An added benefit is that the power circuit cannot crisscross over the small signal circuit.
This IS an upcoming group project, at some point after I understand the triple version of it (for my pair of rather difficult speakers), and have audio amp appropriate bridging methods prepared.
Why do it?
Parallel LM1875 sounds lavish, as if you worked really hard on a discrete amp.
The problem:
The project is indefinitely held up over sound field size (low resistor values) versus a favorable tone (high resistor values), because the two necessary features seem to be mutually exclusive when you try to do it with resistor values alone.
Also, the amp ("board"), if running without any problems, should be able to use 220uF power caps rather than use 470uF to cover up tone problems. Reducing that 470uF value along with running clearer, level, and cooler, would indicate that the amp is ready for prime time.
One odd thing is the DC performance, which is always going to be better with a 10k input load than 100k input load. However, the 10k+feedback shunt+stopper constitutes resistor comp. For example 10k input load, 475R stopper, 1k feedback shunt = 11.47k resistor compensation. That's really very mild, but if carried to a greater extreme, could sharpen up a dull amp or even cause ringing. Its probably great for dull inverting amplifiers. But for some non-inverting amplifiers, it seems that some of these resistor values need to increase.
The source device will probably perform a lot better (as in much better bass) with a 10k load on it, so I wouldn't favor harming that.
Perhaps the 68k feedback resistor has no effect on sound whatsoever.
And that leaves this. . .
Probably the 2.7k feedback-shunt resistor and the 1k stopper, have decreased the amount of resistor comp and avoided actuance in the non-inverting amplifier, indeed a problem that plagues National Semiconductor chips, and perhaps my preference for sky high resistor values and huge gain is reducing that problem? My favorite LM1875 has 114k/2.7k. It doesn't require extra huge power caps point blank on the chip to level it out, but rather plays along with a perfectly level frequency response and. . . 220uF power caps. It runs cool, very clear, and very sturdy, much unlike the datasheet example application.
Instead of my gain comp, your cap comp could give a similar performance and probably better performance. Did I guess right?
From the 114k/2.7k example, I would desire to pull a little more current through there, but without the "hard sound" problem. It seems that 68k/2.2k would have the desired (bigger) effect on sound field size, but the tone inferior to 114k/2.7k.
From reading your post, I believe that two little picofareds small capacitors could make up that difference in tone.
Have your cake and eat it too? Is that what you said?
Will that get me unstuck?
I'm not skilled in math, but for me, examples can usually explain the figures. My grip on the topic is still infirm, but am possibility beginning to understand.
While the ear cannot detect some sounds that are overall too quiet, if instead two given samples are loud enough to hear them, it can indeed detect very tiny harmonic differences as small as 0.03db. A temperature probe on the heatsink, gathering average temperature data, can tell the rest of the story, whereby average cooler is better stability. Clearer and cooler is the win.
PCB? There's only 5 pins for solo or 10 pins for parallel. No need to make it harder with a PCB.
The 2-chip parallel amps are especially good because its a simple, fast, rigid assembly that cannot easily spin off from the insulator pad. Simply bend the v- almost to the chip(s) face(s) and then bend the v+ pins(s) half that distance. Set a 35v low impedance 470uF in from the left, repeat from the right. The ecap pins become the rails, horizontally across the chips. Add 100nF caps to stabilize the "free" ends, one on the left and one on the right. It is expediently fast and simple to do. An added benefit is that the power circuit cannot crisscross over the small signal circuit. This IS an upcoming group project, at some point after I understand the triple version of it (for my pair of rather difficult speakers), and have audio amp appropriate bridging methods prepared.
Why do it?
Parallel LM1875 sounds lavish, as if you worked really hard on a discrete amp.
The problem:
The project is indefinitely held up over sound field size (low resistor values) versus a favorable tone (high resistor values), because the two necessary features seem to be mutually exclusive when you try to do it with resistor values alone.
Also, the amp ("board"), if running without any problems, should be able to use 220uF power caps rather than use 470uF to cover up tone problems. Reducing that 470uF value along with running clearer, level, and cooler, would indicate that the amp is ready for prime time.
One odd thing is the DC performance, which is always going to be better with a 10k input load than 100k input load. However, the 10k+feedback shunt+stopper constitutes resistor comp. For example 10k input load, 475R stopper, 1k feedback shunt = 11.47k resistor compensation. That's really very mild, but if carried to a greater extreme, could sharpen up a dull amp or even cause ringing. Its probably great for dull inverting amplifiers. But for some non-inverting amplifiers, it seems that some of these resistor values need to increase.
The source device will probably perform a lot better (as in much better bass) with a 10k load on it, so I wouldn't favor harming that.
Perhaps the 68k feedback resistor has no effect on sound whatsoever.
And that leaves this. . .
Probably the 2.7k feedback-shunt resistor and the 1k stopper, have decreased the amount of resistor comp and avoided actuance in the non-inverting amplifier, indeed a problem that plagues National Semiconductor chips, and perhaps my preference for sky high resistor values and huge gain is reducing that problem? My favorite LM1875 has 114k/2.7k. It doesn't require extra huge power caps point blank on the chip to level it out, but rather plays along with a perfectly level frequency response and. . . 220uF power caps. It runs cool, very clear, and very sturdy, much unlike the datasheet example application.
Instead of my gain comp, your cap comp could give a similar performance and probably better performance. Did I guess right?
From the 114k/2.7k example, I would desire to pull a little more current through there, but without the "hard sound" problem. It seems that 68k/2.2k would have the desired (bigger) effect on sound field size, but the tone inferior to 114k/2.7k.
From reading your post, I believe that two little picofareds small capacitors could make up that difference in tone.
Have your cake and eat it too? Is that what you said?
Will that get me unstuck?
I'm not skilled in math, but for me, examples can usually explain the figures. My grip on the topic is still infirm, but am possibility beginning to understand.
Last edited:
Guessing is the least efficient method. It is not likely that you guess the right values. Don’t be fooled by people who claim to have found by guessing and listening that certain snubber values in a power supply lead to a better sounding amp. That is only marketing.
An amplifier can be unconditionally stable, conditionally stable or unstable. There are no intermediate states, like 58 % stable is better than 59 % stable. You want either an unconditionally stable amp or an amp that is conditionally stable as long as it is used appropriately. Increased heat dissipation may indicate oscillation, but the opposite is not a reliable proof of the absence of oscillation.
The range from about 15 kHz upward is responsible for this impression. Increasing the volume of that frequency range provides more “air” i.e. appears to create a bigger sound-field, while decreasing that range makes the sound more pleasant.
By increasing the gain, you create a lower frequency roll-off, and by that affect exactly the frequency range above 15 kHz. So, yes, they are mutually exclusive and you can only choose the compromise that suits you best. If you plan on making that a group project, you should however consider that those differences may be much smaller than the ones from one speaker to another. You will need to find the right compromise for different speakers, not only for yours.
What is DC performance?
I assume that resistor compensation is supposed to mean that both inputs see the same impedance to ground. From a DC point of view the amp cannot see what you call the feedback shunt resistor, because it is in series with a capacitor which blocks the path for DC. The inverting input sees ground through the feedback resistor + load. The non-inverting input sees ground through the input resistor + stopper resistor. A stopper resistor is not necessary for the LM1875. If you want the same DC impedance, you choose the input resistor and the feedback resistor to be of the same value.
The cap is primarily there for stability. It can be abused to shape the HF roll-off for a low gain to make it similar to the natural HF roll-off with high gain. Alternatively you could use a corresponding HF filter at the amp input.
Here is a picture.
An amplifier with a phase turn of 135° or more at unity gain is considered unstable. Less than 135° are considered stable. Conditionally stable means that e.g. a high capacitance at the output can increase the phase turn of an amp so that it becomes unstable, although it is stable on the condition that the load is resistive.
The upper graph shows an amp compensated with 22 pF in parallel to the feedback resistor and 220 pF between the input legs. The -3 dB point is around 120 kHz and consequently the amp is 0,1 dB down at 20 kHz. Phase at unity gain (frequency where the gain crosses 0 dB, here about 2,4 MHz) is around -118°. The amplifier is stable and has a margin of 17° to deal with capacitive loads.
The lower graph shows the same amp uncompensated. The -3 dB point is around 400 kHz. The amp is down only 0,01 dB at 20 kHz, but at unity gain (around 7,4 MHz) you see that the phase turn is -141 ° which means this amp is not stable and very likely to oscillate.
So if your amp oscillates those capacitors can cure that at the price of earlier HF roll-off and as usual, the bigger the capacitor, the lower is the roll-off frequency.
Thank you again.
I'm aware that snubber values cannot work except for the specific model of transformer, layout and situation for which they are specific.
I have been able to make use of RMAA; however, I do also find that some of its reports are about ten times understated. . . because its either ten times off or else I can hear 0.03db. One of those is true. I do not know which. However, errors detected are always at the same frequency and in the same direction.
A more useful thing to do with the computer, but I don't know if its available, would be to play a square into itself and zero out any variance from perfect, and then next enclose the amplifier into that loop and look at the square wave difference. Is there not yet some incredibly simple way to do this?
HI Pacific @ Post 131.
That is a well written very useful informative post not that im a description expert by any means. Andrew t understands it although he has much much more amplifier design knowlage than me. For me to understand nearly every word Is amazing and you have certainley cleared up an area i was having a few problems with.. THANKYOU.
Regards Ian
That is a well written very useful informative post not that im a description expert by any means. Andrew t understands it although he has much much more amplifier design knowlage than me. For me to understand nearly every word Is amazing and you have certainley cleared up an area i was having a few problems with.. THANKYOU.
Regards Ian
Yes, there is. You need a square wave generator and an oscilloscope. For good results you will need solutions with additional hardware. The sound card's upper frequency limit is usually OK to use purely software-based (even freeware) signal generators and oscilloscopes for speaker design, yet too restricted for thorough amplifier diagnostics.
I have been through this entire thread and also have written BrianGT four times and gotten no response. No links to any docs seem to work. I have the Chipamp 1875 kit. But I would like to get a schematic, parts list and possibly some instructions before assembling it. Does anyone know where I can find these? I find it unbelievable that Brian would sell these kits without these and furthermore would completely ignore all communications regardng the kits he sells. This is not cool at all. I'd appreciate getting help from anyone at all at this point. Thanks much.
Hello Steve,
The parts list is here: LM1875 Amplifier Kit | Chipamp.com
The LM1875 datasheet will give you the diagram
My personal opinion is that point to point wiring is much easier.
Jim
The parts list is here: LM1875 Amplifier Kit | Chipamp.com
The LM1875 datasheet will give you the diagram
My personal opinion is that point to point wiring is much easier.
Jim
Hello Steve,
Well it can't be far off the datasheet they never are
Here's a pic of my latest effort p2p around the chip as I said above;
Jim
Well it can't be far off the datasheet they never are
Here's a pic of my latest effort p2p around the chip as I said above;
An externally hosted image should be here but it was not working when we last tested it.
Jim
Thanks again Jim. I see point to point is certainly doable. But I'll use the boards this time. I suppose the boards are simple enough that I can trace the lines and accurately reconstruct the schematic to be sure it follows the NatSemi data sheet. But I was hoping to avoid that. Either way I think I have enough data at this point to have a go at building the amp.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.