Hello all,
I was wondering what the input impedance of an op-amp like LM4562 would be, in the absence of a power supply (0V). My wild guess is that it would still be high (not sure though), as the input stage biasing would be lost, deactivating the differential pair, muting the amplifier. Please correct me if you think I'm wrong.
The basis for the question is the idea of a simple attenuation ON/OFF switch that would short out the power supply of an inverting attenuator (using MOSFET/relay etc.), to turn off the attenuation (picture). After removal of power, the signal path to the next stage (power amp) is expected to be retained through the resistors Ri and Rf.
Please note that I would like to avoid using controls / switches or digital attenuation for varying the gain, as the number of channels is just too many (16). Looking forward to reading your comments. Thank you.
I was wondering what the input impedance of an op-amp like LM4562 would be, in the absence of a power supply (0V). My wild guess is that it would still be high (not sure though), as the input stage biasing would be lost, deactivating the differential pair, muting the amplifier. Please correct me if you think I'm wrong.
The basis for the question is the idea of a simple attenuation ON/OFF switch that would short out the power supply of an inverting attenuator (using MOSFET/relay etc.), to turn off the attenuation (picture). After removal of power, the signal path to the next stage (power amp) is expected to be retained through the resistors Ri and Rf.
Please note that I would like to avoid using controls / switches or digital attenuation for varying the gain, as the number of channels is just too many (16). Looking forward to reading your comments. Thank you.
+1 Nonlinear and non-predictable.
The only thing I would predict is that you likely get a diode clamp at the input due to the ESD structures within the chip. That'll limit the input voltage at the pin so somewhere around 0.5-0.7V.
If you want to make a switchable attenuator, why not just have a relay that bypasses the attenuator? Or controls the gain if you're looking for switched gain settings.
Tom
The only thing I would predict is that you likely get a diode clamp at the input due to the ESD structures within the chip. That'll limit the input voltage at the pin so somewhere around 0.5-0.7V.
If you want to make a switchable attenuator, why not just have a relay that bypasses the attenuator? Or controls the gain if you're looking for switched gain settings.
Tom
Almost all IC's have input protection diodes to V+ and V-. As such things get very nonlinear. IC's can actually become powered up by by the input signal. This can become very confusing especially with digital IC's and be difficult to debug.Other conduction paths depending on the input pair type and topology may also become involved. Don't expect the input source to to not see peculiar loading. At least with the inverting topology you have chosen the Zin will not fall below Ri.
Thank you everybody for the responses. I am not very good with the insides of ICs, but I guess the diodes you mention are between + & - inputs (differential clamp), with ESD diodes from each pin to V+ & V-, correct? Now I can see what I was missing.
Going that way, all channels would have to controlled together, with as many switches.
Since the power supply idea is almost guaranteed to fail, I could try CMOS quad switches like 74HC4066. But are there any tricks to linearise them so that they could be used as part of Ri or Rf, to switch gains as mentioned above? Or should I use them to bypass the attenuation circuit somehow?
Going that way, all channels would have to controlled together, with as many switches.
Since the power supply idea is almost guaranteed to fail, I could try CMOS quad switches like 74HC4066. But are there any tricks to linearise them so that they could be used as part of Ri or Rf, to switch gains as mentioned above? Or should I use them to bypass the attenuation circuit somehow?
Last edited:
Based on your comments above, I came up with the switched gain circuit in the picture below. For a 10V (+/-5V) supply, the non-linear error per switch is supposed to be ~5 ohms, while the off-state leakage is below -100dB at 20kHz, as per the 74HC4066 datasheet.
Although this one could work, I'm still not sure about its performance. Do you people see any issues relating to noise, distortion etc. with this circuit? Thanks.
Although this one could work, I'm still not sure about its performance. Do you people see any issues relating to noise, distortion etc. with this circuit? Thanks.
Last edited:
Yes and thank you very much, for a beautiful circuit from times when Gilbert cells were considered expensive. Considering that their function is mostly only multiplication, they remain rather expensive (vs. programmable uCs and DSPs) even today. A volume control would be something very similar, only discrete-step, logarithmic and inexpensive!
After post #8, I was trying to restrict the switch currents to the negligible input current of the opamp, so that their resistance variations do not show up in the transfer functions. However, after reading your article, I am beginning to feel that this current would cut right through the OFF-state parasitic capacitances of most modern switches (not sure though), also possibly giving the circuit in #7 an advantage.
Sir, my exact issue is the following. I have ~2V peak full scale from a DAC that I want to send to a power amplifier. Unfortunately, as most chip-amps being unity gain unstable, they're set for gains higher than 20 or 26 dB (as you quite possibly already know). I calculate only a +6dB gain for the power amplifier, based on reference SPL values (105dB peak) at my listening distance, that brought me to a selectable -20dB attenuation. However, to add to all this, I have some 16 channels that need to be controlled simultaneously. I do not like the digital attenuation methods that keep losing a bit every -6dB. I am OK even with a fixed ratio, in case gain switching is not worth it, please advise. Nevertheless, I'm still happy to learn something. Thank you, once again, for your article on the topic.
Regards.
Last edited:
Well you could do with a fixed 6dB attenuator. Or you could just turn down your level control a bit.
But if you want to go all the way, you need a CS3308 or -18 https://www.mouser.be/datasheet/2/76/CS3308_F1-1141248.pdf
I've used those with excellent results, they are optimized for what you want. Use two for 16 channels, all in lock-step, and the option to finetune each channel and add balance functionality.
See DCX2496 active output mod & 6-channel vol control | Linear Audio NL
And to head off some comments: no, it is NOT a digital attenuator. It is an analog attenuator, switched resistor attenuator, digitally controlled ;-)
Jan
But if you want to go all the way, you need a CS3308 or -18 https://www.mouser.be/datasheet/2/76/CS3308_F1-1141248.pdf
I've used those with excellent results, they are optimized for what you want. Use two for 16 channels, all in lock-step, and the option to finetune each channel and add balance functionality.
See DCX2496 active output mod & 6-channel vol control | Linear Audio NL
And to head off some comments: no, it is NOT a digital attenuator. It is an analog attenuator, switched resistor attenuator, digitally controlled ;-)
Jan
Last edited:
Thank you, I I'll use a fixed one, as you said. I was worrying about a high attenuation factor like -20dB causing issues with fidelity, when done digitally. I already have digital (soft) volume on the processor (not Behringer though) and DAC in small steps for all channels (24-bit), wherein bit loss may not be significant.
Meanwhile, reading your article inspired me to come up with something similar (attached), although it may not be suitable for applications (such as mine) that have a high channel count. It was the NE5532 that was used in the simulation, but I guess a FET-input opamp like OPAx134 with a current noise density in the fA/√Hz would work better (I maybe wrong). With three stages, the maximum attenuation is past 140dB in 1dB steps, indeed a nice exercise during lockdown.
View attachment Volume control.pdf
Meanwhile, reading your article inspired me to come up with something similar (attached), although it may not be suitable for applications (such as mine) that have a high channel count. It was the NE5532 that was used in the simulation, but I guess a FET-input opamp like OPAx134 with a current noise density in the fA/√Hz would work better (I maybe wrong). With three stages, the maximum attenuation is past 140dB in 1dB steps, indeed a nice exercise during lockdown.
View attachment Volume control.pdf
Last edited:
Yes that will work, but as you say you have three opamps in series now. And you probably don't need 1dB steps, even 3dB is barely noticeable. And more attenuation than 60 or 70dB also not really needed, so you probably could optimize it with one stage with 3dB steps and 60dB range, possible with a mute switch.
At this point, you're in the world of stepped attenuators and the 'net is littered with projects for this, as well as here at diyaudio.
Logarithmic Attenuator Calculator
Resistor-Switch Attenuator Networks for Audio Volume Control
Important point is to have fun though!
Jan
At this point, you're in the world of stepped attenuators and the 'net is littered with projects for this, as well as here at diyaudio.
Logarithmic Attenuator Calculator
Resistor-Switch Attenuator Networks for Audio Volume Control
Important point is to have fun though!
Jan
Last edited:
Yes Sir, even I was thinking something like that and just calculated a -114dB attenuator using a single amplifier (attached). I agree with your opinion that any levels below -60dB are not at all required for volume controls, as they're barely audible to many. I will try and modify this 114dB circuit for 3dB steps and it would (maybe) reach -57dB or so.
View attachment 114dB single amplifier.pdf
View attachment 114dB single amplifier.pdf
Yes, understood. I had noticed it when I was going through the links in #16. But isn't that easily overcome by driving the circuit using another amplifier?
I also happened to notice that the said methods give only one gain position per switch, needing as many switches as positions, due to non-utilisation of switch combinations. Besides, occupying more board space would only make larger loops around the ICs, further complicating matters. However, IC implementations would be exceptions, as the geometries are much much smaller than what could be had on PCBs.
Here is the 3dB derived circuit (attached), using a single stage and 8 bilateral switches as before. Maybe 60dB could be called "mute" or something. However, it's nice to see that there are no megaohm resistors at play if one steers clear of anything below -60dB.
View attachment 57dB in 3dB steps.pdf
I also happened to notice that the said methods give only one gain position per switch, needing as many switches as positions, due to non-utilisation of switch combinations. Besides, occupying more board space would only make larger loops around the ICs, further complicating matters. However, IC implementations would be exceptions, as the geometries are much much smaller than what could be had on PCBs.
Here is the 3dB derived circuit (attached), using a single stage and 8 bilateral switches as before. Maybe 60dB could be called "mute" or something. However, it's nice to see that there are no megaohm resistors at play if one steers clear of anything below -60dB.
View attachment 57dB in 3dB steps.pdf
newvirus2008 said:
I was wrong, as this is true only for the string/rotary ladder method. The modified R-2R ladder effectively utilises all combinations and its switch count is exactly equal to the base-2 logarithm of the number of positions, like in a DAC.
However, each switch is of the double-throw type, essentially made of two single-throw switches. Thus, the R-2R with 4 nos. of SPDT switches has 16 positions, whereas the circuit in #17 reaches 20 positions using 8 nos. of SPST ones.
Nevertheless, when compared to the rather straightforward rotary/string ladder, the R-2R method looks like the "more proper" method for stepped attenuation.
jan.didden said:
Yes, that's appears to be a disadvantage, but at least the source-loading problem could possibly be alleviated by using an input buffer, I guess. The noise remains.