Replace "often" with "may" and I would almost agree. "Almost" because its not the feedback principle or topology which is evil here, its the slow base amp. Otherwise said, it is not possible to have a 100% TIM free base amp, and by closing the feedback loop to suddenly have lots of TIM. I'm of course talking about passive/wideband feedback loop circuitry. In fact, in general, the feedback TIM performance is controlled by the loop gain performances. The other way around is that you can have a base amp with some TIM and, by applying feedback, to reduce TIM.
See, Curly? That's how myths are propagated. Lacking a good understanding of these (not always obvious) facts, you are pushed to "trust" one source or another. And what would be your best bet? Either those sources that fit your interests, those "gurus" having a more or less vested interest in distorting the reality, or the self appointed "experts". One of the ground rules of trust (beyond some common sense) should be "follow the money" (or avoid, thereof). Of course, your best bet is to get your hands (and brain) dirty and reach the conclusions by yourself.
To answer your question, no, feedback is not necessary always good. Relying of feedback to correct an otherwise crappy design is flat wrong, for a number of good reasons. It is always a good idea to linearize as much as possible the base amp, then add feedback, but not more than its required to preserve the system stability. OTOH, closing the loop always requires a special treatment of the base amp (not necessary related to linearity) called frequency compensation. I dare to say that the lack of understanding of the frequency compensation intricacies (perhaps beyond the very basic Miller) is what made so many self appointed "experts" to blame feedback.
So do I, in particular uA741, LM301 and other similar from the 60s.
Since then, technology has advanced "a little", just enough to make your then experience and knowledge obsolete. Like it or not, the "halving time" of an EE nowadays is less than 5 years.
We went through this exercise over, and over, and over. As long as you are unable or unwilling to write down the specification of a circuit that you consider "good enough", this type of exercise is already proven as useless.
My offer is still in place: write down the spec of an audio gain stage (or voltage regulator) that you are happy with, and you already built it with discretes, and I'll show you how to implement the same spec or better, more or less exclusively with opamp(s).
But even so, I'm afraid the discussion will quickly slip towards "how it sounds" where everything breaks down to nonsense.
BTW, in my dayjob business the halving time of an engineer is one year.
Ok, here are a few considerations around the head amp concept, in particular about a bipolar implementation.
A head amp is a very low noise, wideband and as much as possible linear amplifier that connect between the MC or MM cartridge and the RIAA correction circuitry. For all purposes, the head amp sets the overall Signal to Noise ratio (S/N) of the entire audio chain, from the vinyl record to the speakers.
Why a head amp? The main reason for using a head amp is because it's difficult to combine a gain stage with a precise RIAA correction, while maintaining a good degree of linearity.
It is usually considered that a good MC or MM cartridge has about 0.1% distortions. Then why strive for, say, better than 0.01% distortions in the amp and not use an open loop design and passive RIAA correction that can easily reach this number at the nominal input level? Main reason is the dynamic range.
MC/MM cartridges output level is usually defined at 1KHz. A 0.5mV MC cartridge playing a record will though output 0.05mV at 100Hz and no less than 5mV at 20KHz. Reason for this is the RIAA correction used to cut the vinyl. OTOH, assuming a 60dB gain at 1KHz, the RIAA amp gain will be 80dB at 100Hz and only 40dB at 20KHz. As you see, we got from the beginning a 40dB dynamic range requirement.
Now, add to that the dynamic range required for the program itself, and some reserve for the vinyl "pops" recovery (you don't want the "pops" to clip the amp and make the clipping recovery impact the sound) and you'll get to the point to which the dynamic range of a RIAA amp should be huge. Tubes are excellent here; however we are talking solid state here. Therefore, there's a trade between the dynamic range (aka "headroom") and the practical considerations for building such a SS amp.
Of course, the headroom translates in a maximum output swing before clipping, which translates into a maximum input level. It is very interesting to note that it is not enough to consider the headroom at a single frequency. The headroom may significantly vary with the frequency (because different stages may have different frequency responses) therefore clipping doesn't necessary occur in the output stage. This is in particular important when the head amp concept is used, as it will be eventually discussed below.
To put some numbers in the picture, consider the HPS 3.1 amp. The output swings at +/-20V peak and the gain is 60dB. Therefore, the maximum allowed input level before clipping is 20mV peak. 20mV/0.5mV = 40, or 32dB. This is the HPS3.1 headroom. Of course, what is assumed in this very simple calculation is that the max output is +/-20V peak at all frequencies. A great deal of design effort was put in providing such a feature, as it will be eventually described below. As a rule, providing a combination of high gain and high dynamic range/headroom is increasingly difficult, and the product of such, divided by the S/N, can be considered as a good figure of merit (or metric) for a certain RIAA amp.
Now, back to the distortions, I would certainly agree that providing 0.01% distortions for a 0.5mV input by using an open loop design is very easy. Not so easy for 20mV, unless some sort of degeneration is used (or a multi-tanh scheme for bipolars). Degeneration always adds noise, while multi-tanh is very difficult to apply in discrete designs (which are currently unavoidable for designs with under 1nV/rtHz noise). As a rule, as soon as the input approaches the thermal voltage kT/q=26mV, the distortions start increasing exponentially. It is certainly a challenge to design a 0.01% distortion open loop gain stage, for all frequencies (say 20Hz-50KHz) and input levels from 0.5mV to 20mV or more. If you consider that some designs are reusing the same input stage for MM cartridges (with one order of magnitude larger outputs, therefore a 200mV maximum input) this becomes practically impossible.
Therefore, the concept of head amp and the idea (actually not really new) to close a feedback loop around the low noise input stage, to help linearizing the input stage for a wide range of inputs.
Now, what's the price of using a head amp? There's a pretty hefty price to pay, which comes exactly from the same dynamic range department. By its nature, a head amp has a constant gain across the entire audio frequencies range. The headamp can't be made to high, because it would have the same gain at 20KHz, where the input is 20dB higher. True, the following RIAA stages will have -20dB gain at 20KHz, but there's the risk that clipping will occur in the head amp, before the output stage clips. OTOH, the head amp could not be smaller than 20dB, because otherwise the RIAA stages noise couldn't be neglected, and the head amp setting the S/N requirement would be violated. Some thinking, and some algebra, would show that an optimum is reached when the head amp gain equals the required dynamic range. From now on, all that needs to be determined is the supply voltage, to allow the maximum required output swing. That's the reason why in HPS3.1 and HPS4.0 the head amp gain was set to 40 (32dB), equal to the entire RIAA amp headroom.
Back to the feedback loop, for a conventional resistive feedback network, there's always a noise impact. Current feedback is the least daunting, because the degeneration (required to linearize the open loop characteristic) can be made part of the feedback network. In HPS3.1 and HPS4.0 four 1 ohm resistors are providing degeneration/local feedback (to a noise equivalent of 1/4~0.25ohm per stage). For a fixed gain of 40, the other feedback resistor (that does not directly impact the noise, because its noise equivalent is divided by the open loop gain) results 39ohm.
The output level at the head amp will be 0.5mV*40 = 20mV at 1KHz. At 20KHz it will be maximum 200mV. Add the headroom of 40 (32dB) and you'll find the worst case output level as being not less than 8V. See how things are adding up? We started with on 0.5mV nominal signal and we got to an 8V maximum output!
These 8V are fully dropping on the 39ohm feedback resistor. This means that the head amp has to provide no less than 200mA of current to be able to sustain this output level. Of course, this is an absolute maximum rating, the average current and power dissipation will be much lower.
Now, about the bipolar version.
Bipolar devices are quieter than unipolar (JFET or MOSFET) devices. It is well known though that bipolar devices have both significant equivalent input voltage and current noise, while in unipolar devices, at least for RIAA amps, the input current noise can be safely neglected.
Obviously, a bipolar head amp can be designed to have much less noise than a JFET head amp. Also, low noise bipolar devices are still widely available (and I would mention here the Sanyo FBET devices, at par with the best dedicated low noise devices ever made). What are the challenges of a bipolar head amp?
There are a few. First, the murky cartridge current. I was unable to find out what is the maximum DC current allowed by a cartridge, and what is the impact on the AC performance. The automatic answer is "zero current", unfortunately this is not really possible to achieve in a bipolar implementation. Even in a complementary symmetrical design (like HPS4.0), beta mismatches will lead to a DC current through the DC coupled cartridge. I was able to bring this current to 0.5uA, using selected and matched high beta devices, but I would agree this is difficult and impractical to target for DIYers.
Now, assuming the fear of DC current is based on permanently magnetizing the cartridge, an input cap of whatever quality won't help a iota. As long as the cartridge impedance is much lower than the amp input impedance, a relatively large current pulse will flow anyway at power up or in any other transient conditions; if anything bad can happen, it will happen with or without a $100 teflon isolation cap, not to mention a leaky electrolytic. It appears that, like it or not, we have to live with some DC through the cartridge.
Secondly, from a noise perspective, there's an optimum in the collector current of the low noise input stage. This is Ic=Vt*SQRT(Beta)/Rsource. Assuming Vt=26mV, Beta=100 and Rsource=10ohm, Ic=26mA and the base current is 260uA. At this levels, it is extremely difficult to balance the input bias currents to under 1uA. Of course, it is always possible to use a input bias current cancelling technique, but it can be easily proven from thermodynamic considerations that input bias current cancellation can't be done without a noise penalty.
Third, the input impedance trade-off. Input impedance can be made quite high, by using negative feedback, and/or emitter followers at the input. It is not a huge difficulty to bring it in the 47kohm range, which seems to be the standard for high input impedance. However, the input bias current is biting us again. If we use a 47k input resistor to the ground, the input offset voltage (that cannot be cancelled by a similar resistor on the inverting input, because that has to be low as part of the feedback network, see above) can easily reach 500mV (50k*10uA), in particular over a large range of temperatures. 500mV*40=20V output and our head amp is dead in the water. No servo could compensate such a huge offset.
I'll continue ASAP... Hope this will explain the philosophy around the head amp concept, and the design strategy, that I am using in the HPS 3.1 and up RIAA amps designs.
A head amp is a very low noise, wideband and as much as possible linear amplifier that connect between the MC or MM cartridge and the RIAA correction circuitry. For all purposes, the head amp sets the overall Signal to Noise ratio (S/N) of the entire audio chain, from the vinyl record to the speakers.
Why a head amp? The main reason for using a head amp is because it's difficult to combine a gain stage with a precise RIAA correction, while maintaining a good degree of linearity.
It is usually considered that a good MC or MM cartridge has about 0.1% distortions. Then why strive for, say, better than 0.01% distortions in the amp and not use an open loop design and passive RIAA correction that can easily reach this number at the nominal input level? Main reason is the dynamic range.
MC/MM cartridges output level is usually defined at 1KHz. A 0.5mV MC cartridge playing a record will though output 0.05mV at 100Hz and no less than 5mV at 20KHz. Reason for this is the RIAA correction used to cut the vinyl. OTOH, assuming a 60dB gain at 1KHz, the RIAA amp gain will be 80dB at 100Hz and only 40dB at 20KHz. As you see, we got from the beginning a 40dB dynamic range requirement.
Now, add to that the dynamic range required for the program itself, and some reserve for the vinyl "pops" recovery (you don't want the "pops" to clip the amp and make the clipping recovery impact the sound) and you'll get to the point to which the dynamic range of a RIAA amp should be huge. Tubes are excellent here; however we are talking solid state here. Therefore, there's a trade between the dynamic range (aka "headroom") and the practical considerations for building such a SS amp.
Of course, the headroom translates in a maximum output swing before clipping, which translates into a maximum input level. It is very interesting to note that it is not enough to consider the headroom at a single frequency. The headroom may significantly vary with the frequency (because different stages may have different frequency responses) therefore clipping doesn't necessary occur in the output stage. This is in particular important when the head amp concept is used, as it will be eventually discussed below.
To put some numbers in the picture, consider the HPS 3.1 amp. The output swings at +/-20V peak and the gain is 60dB. Therefore, the maximum allowed input level before clipping is 20mV peak. 20mV/0.5mV = 40, or 32dB. This is the HPS3.1 headroom. Of course, what is assumed in this very simple calculation is that the max output is +/-20V peak at all frequencies. A great deal of design effort was put in providing such a feature, as it will be eventually described below. As a rule, providing a combination of high gain and high dynamic range/headroom is increasingly difficult, and the product of such, divided by the S/N, can be considered as a good figure of merit (or metric) for a certain RIAA amp.
Now, back to the distortions, I would certainly agree that providing 0.01% distortions for a 0.5mV input by using an open loop design is very easy. Not so easy for 20mV, unless some sort of degeneration is used (or a multi-tanh scheme for bipolars). Degeneration always adds noise, while multi-tanh is very difficult to apply in discrete designs (which are currently unavoidable for designs with under 1nV/rtHz noise). As a rule, as soon as the input approaches the thermal voltage kT/q=26mV, the distortions start increasing exponentially. It is certainly a challenge to design a 0.01% distortion open loop gain stage, for all frequencies (say 20Hz-50KHz) and input levels from 0.5mV to 20mV or more. If you consider that some designs are reusing the same input stage for MM cartridges (with one order of magnitude larger outputs, therefore a 200mV maximum input) this becomes practically impossible.
Therefore, the concept of head amp and the idea (actually not really new) to close a feedback loop around the low noise input stage, to help linearizing the input stage for a wide range of inputs.
Now, what's the price of using a head amp? There's a pretty hefty price to pay, which comes exactly from the same dynamic range department. By its nature, a head amp has a constant gain across the entire audio frequencies range. The headamp can't be made to high, because it would have the same gain at 20KHz, where the input is 20dB higher. True, the following RIAA stages will have -20dB gain at 20KHz, but there's the risk that clipping will occur in the head amp, before the output stage clips. OTOH, the head amp could not be smaller than 20dB, because otherwise the RIAA stages noise couldn't be neglected, and the head amp setting the S/N requirement would be violated. Some thinking, and some algebra, would show that an optimum is reached when the head amp gain equals the required dynamic range. From now on, all that needs to be determined is the supply voltage, to allow the maximum required output swing. That's the reason why in HPS3.1 and HPS4.0 the head amp gain was set to 40 (32dB), equal to the entire RIAA amp headroom.
Back to the feedback loop, for a conventional resistive feedback network, there's always a noise impact. Current feedback is the least daunting, because the degeneration (required to linearize the open loop characteristic) can be made part of the feedback network. In HPS3.1 and HPS4.0 four 1 ohm resistors are providing degeneration/local feedback (to a noise equivalent of 1/4~0.25ohm per stage). For a fixed gain of 40, the other feedback resistor (that does not directly impact the noise, because its noise equivalent is divided by the open loop gain) results 39ohm.
The output level at the head amp will be 0.5mV*40 = 20mV at 1KHz. At 20KHz it will be maximum 200mV. Add the headroom of 40 (32dB) and you'll find the worst case output level as being not less than 8V. See how things are adding up? We started with on 0.5mV nominal signal and we got to an 8V maximum output!
These 8V are fully dropping on the 39ohm feedback resistor. This means that the head amp has to provide no less than 200mA of current to be able to sustain this output level. Of course, this is an absolute maximum rating, the average current and power dissipation will be much lower.
Now, about the bipolar version.
Bipolar devices are quieter than unipolar (JFET or MOSFET) devices. It is well known though that bipolar devices have both significant equivalent input voltage and current noise, while in unipolar devices, at least for RIAA amps, the input current noise can be safely neglected.
Obviously, a bipolar head amp can be designed to have much less noise than a JFET head amp. Also, low noise bipolar devices are still widely available (and I would mention here the Sanyo FBET devices, at par with the best dedicated low noise devices ever made). What are the challenges of a bipolar head amp?
There are a few. First, the murky cartridge current. I was unable to find out what is the maximum DC current allowed by a cartridge, and what is the impact on the AC performance. The automatic answer is "zero current", unfortunately this is not really possible to achieve in a bipolar implementation. Even in a complementary symmetrical design (like HPS4.0), beta mismatches will lead to a DC current through the DC coupled cartridge. I was able to bring this current to 0.5uA, using selected and matched high beta devices, but I would agree this is difficult and impractical to target for DIYers.
Now, assuming the fear of DC current is based on permanently magnetizing the cartridge, an input cap of whatever quality won't help a iota. As long as the cartridge impedance is much lower than the amp input impedance, a relatively large current pulse will flow anyway at power up or in any other transient conditions; if anything bad can happen, it will happen with or without a $100 teflon isolation cap, not to mention a leaky electrolytic. It appears that, like it or not, we have to live with some DC through the cartridge.
Secondly, from a noise perspective, there's an optimum in the collector current of the low noise input stage. This is Ic=Vt*SQRT(Beta)/Rsource. Assuming Vt=26mV, Beta=100 and Rsource=10ohm, Ic=26mA and the base current is 260uA. At this levels, it is extremely difficult to balance the input bias currents to under 1uA. Of course, it is always possible to use a input bias current cancelling technique, but it can be easily proven from thermodynamic considerations that input bias current cancellation can't be done without a noise penalty.
Third, the input impedance trade-off. Input impedance can be made quite high, by using negative feedback, and/or emitter followers at the input. It is not a huge difficulty to bring it in the 47kohm range, which seems to be the standard for high input impedance. However, the input bias current is biting us again. If we use a 47k input resistor to the ground, the input offset voltage (that cannot be cancelled by a similar resistor on the inverting input, because that has to be low as part of the feedback network, see above) can easily reach 500mV (50k*10uA), in particular over a large range of temperatures. 500mV*40=20V output and our head amp is dead in the water. No servo could compensate such a huge offset.
I'll continue ASAP... Hope this will explain the philosophy around the head amp concept, and the design strategy, that I am using in the HPS 3.1 and up RIAA amps designs.
jan, i look at amplication stages as building blocks so i misst the point that the signal goes into the amplification stage first and is second fed back.
I did not want to put forward an analysis of the physical causes and remedies just wanted to make shure what we are talking about and again i learned slew limmiting is the cause for TIM. Of cause a tube stage with it´s incredibly high voltage margin works only on a small part of it´s posible swing and that gives it good linearity without NFB. When we talk about solid state i think NFB done right is of great help to lower distortion and anhance bandwidth. Why it still has such a bad reputation today is a mystery to me.
To clarify it for me ones and for all i am planning to build two versions of my MPP phonostage, one with lokal currentfeedback and one with overal NFB. Maybe at the tiy input signal that an MC cartridje sends there COULD be a resolution limmit but so far my experience tells me otherwise.
I did not want to put forward an analysis of the physical causes and remedies just wanted to make shure what we are talking about and again i learned slew limmiting is the cause for TIM. Of cause a tube stage with it´s incredibly high voltage margin works only on a small part of it´s posible swing and that gives it good linearity without NFB. When we talk about solid state i think NFB done right is of great help to lower distortion and anhance bandwidth. Why it still has such a bad reputation today is a mystery to me.
To clarify it for me ones and for all i am planning to build two versions of my MPP phonostage, one with lokal currentfeedback and one with overal NFB. Maybe at the tiy input signal that an MC cartridje sends there COULD be a resolution limmit but so far my experience tells me otherwise.
It is difficult to PRECISELY tell anyone the exact, in's and out's of TIM and other related distortions. We have worked on the problem for more than 40 years, and there is a ton of material out there, but most of you here, have neither the wisdom or the ability to sift through it all to get it right.
Syn08 is partly right, Jan is partly right, but only in a 1/2 baked way.
In the end, it is what comes of each design approach and how it fares in listening, as well as measurement.
Syn08 is partly right, Jan is partly right, but only in a 1/2 baked way.
In the end, it is what comes of each design approach and how it fares in listening, as well as measurement.
Not anymore. Last time was the summer of 1997 when I toured a number of US universities (Stanford, U of T in Austin and TAMU come to mind) as a visiting professor. Since then, I switched to the corporate world and never looked back.
I am sure many commercial designs are doing much worse, only they never specify such. As usual, ignorance is bliss.
Last edited:
i think a little bit DC is not a big problem. i pushes the coils a little off center and that creates some 2nd harmonic. One solutions is to simply use a bit less down preasure to bring them back in line. I have worked for years with direct coupled bipolar designs and have never fried a cartridge.
Steve and mlloyd1
I fully agree a very good discussion and presentation by syn08.
I agree in everything he is writing.
About multi-tanh:
The linear region for Vin=kT/q is the rule for a cell with an Aeff=4 (a bit lower)
The linear region can be increased (with a few very small bumps).
An Aeff=4 is possible to make with discrete components.
Anyway the multi-tanh requires a PTAT circuit.
Cheers
I fully agree a very good discussion and presentation by syn08.
I agree in everything he is writing.
About multi-tanh:
The linear region for Vin=kT/q is the rule for a cell with an Aeff=4 (a bit lower)
The linear region can be increased (with a few very small bumps).
An Aeff=4 is possible to make with discrete components.
Anyway the multi-tanh requires a PTAT circuit.
Cheers
Yes, unfortunately practically it is very difficult to keep it all together: linearity, noise and complexity. I am still looking into this option, but at this point I have little hopes to get something workable.
Of course, everything was already done 40 years ago.
And I can paint like Rembrand, here's a sample:
See? Only some details are missing...
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.