Tweet
This statement is from Lowell in his Vox Berleley II discussion:
“Good stuff.... Pretty wild that solid state is so demonized, when its really the CIRCUIT, not using ss devices, that dictates the sound. Inspires me to wanna build a ss amp that could pass a blind fold test. This amp sure as hell does!!”
Solid state components have come a long way since that design; it should be possible to do much better now!
My further interpretation Lowell’s statement: instead of designing solid state circuits in the most convenient way, or minor variations of what everybody usually does, rather one should look and see how a tube amp does what it does and try to achieve a similar funcionality with solid state.
The goal is to design a solid amp equivqlent to a typical tube amp consisting of a high gain preap section and a power section. Another goal is to make the front end much quieter than a tube preamp. The main purpose of this is to allow the guitar volume to be used with good SNR over a wide dynamic range. (That requires some changes to the standard guitar electronics as well, of course.)
Consider the high gain preamp part first. (The power section is not done yet.) A high gain tube preamp uses several independent stages of gain with gain controls in between and a tone stack as well. A block diagram of one realization is given here: FETpreamp.pdf (along with the actual circuit, to be discussed below). So the solid state will do the same, but what to use? Some of requirements are:
1. The first stage must be extremely quiet. But given the flexibility in adjusting levels in the block diagram, really both the first and second stage should be very quiet. Well, in that case, it makes sense to make them all the same, and thus all quiet.
2. We want the device to be sort of tube like in its characteristics, if possibe, or better, have adjutable linearity and limiting characteristics.
So the first two requirements suggest junction FETs: pretty nice curves, and certain modern FETS are one nanovolt per root Hz, or even a bit better. Also the noise is really good noise: it is all voltage noise, almost no current noise, the kind that depends on the impedance connected at the input. This preamp must be good for a standard guitar pickup with cable, and that has a nasty high impedance resonance right where you do not want it from a noise perspective, exaggerating the effect of current noise.
So, if we are to use FETS, the rest of the requirements are concerned with how to use them.
3. The FET has higher capacitances than the tube, and so we must avoid amplifying it at the input (Miller effect).
The quietest circuit is probably a cascode, but
4. Given the square law response of the FET, its linearity is not as good as a triode, and so we need to improve the linearity with a good circuit.
For this application there is a better circuit than cascode (the usual choice to avoid the Miller effect): a source coupled pair. The circuit of this noninverting Stage of Gain (SoG) is shown in the same attachmnet. (The FETs are LSK170F, reasonably well matched for the left and right positions.) The noise voltage is worse by the square root of two since both devices contribute, but it is very linear for small signals. Maybe this circuit can be too linear, and so we want a way of adjusting the linearity as required.
The source coupled FET pair is a remarkable circuit. (Now would be a good time to review EB’s articles on JFETs: http://www.linearsystems.com/assets/...%20Bordely.pdf For a constant drain source voltage, the current varies as the square of the gate source voltage. That is, the transcondance varies linearly with VGS. This means that if two FETs are connnected as a source coupled pair and are matched, then the individual gate source voltages change oppositely, and the variation of the transcondances disappears. This is true even if the quiescent currents through the two FETs are not the same. This results in extremely good linearity, assuming that the FET is very close to a true current source. And it is, as long as VDS is large. However, the output resistance drops as VDS drops, resulting in another source of non-linearity. (See Figure 2A of the first EB paper linked to above.) So this suggests a way of making a preamp circuit with adjustable linearity: vary the biasing so that the quiesent VDS changes.
This is not an amplifier with a differential output. If we feed the signal into the gate of the left FET, its drain must go directly to the power supply (AC ground) in order to have no Miller effect. The output signal is taken from the drain of the right FET, with the usual resistor to the power supply.
The current limiting is excellent with this circuit. When the left FET turns off, the right FET gets all the current, doubling the equlibrium value. When the left FET turns on full, the right FET goes off, and the output is connected to the power supply through the drain resistor. (No power supply rejection in this circuit; we need good bypassing.) Some might remember ECL (emitter coupled logic). There is sort of an anlogy here to the way this circuit limits.
The power supply bypass capacitors are included in the circuit diagram of the SoG, and so this becomes a complete unit that can be duplicated for any of the four locations in the block diagram. In addition to the input and output a third line is brought out. This is the high quality ground reference, but it can also serves as a way to lower VDS (by introducing a small dc voltage), causing an asymmetrical non-linearity, producing even and odd harmonics.
Here are some measured characteristics, taken at the test output, that is, after three stages. First the noise; it is meausred by taking the spectrum with the input grounded, and then again with a calibration source, a 10K metal film resistor. This is not an easy measurement; the power spectra result from averaging many several second long FFTs. Then the calculations are performed to turn noise and noise plus cal into noise voltage. The spectra have power line harmonics and other spikes probably from nearby switching power supplies. They are too small to be heard, but they contaminate the delicate measurement. Thus the result is filtered in frequency with a low pass median filter which suppresses the outliers, giving the smooth measurement shown here:
. The noise voltage is about 1.6 nV/rtHz at high frequencies, rising at lower frequencies as a result of the always present 1/f noise. 1.6 nV/rtHz corresponds to about 151 ohms. Some tube ampa use 68K in series with the grid; adding 2K for the 12AX7 and is almost 27 db more noise than this FET preamp. A tube amp might be only 16 db worse with careful design and construction, still using a 12AX7.
Next some numbers on the linearity with the symmetry adjust on zero volts. With a suitable sine wave fed to the input, the two level controls before the test output are adjusted so that the signal increases from stage to stage. When the outpout at the test output is 300 mv RMS, the highest harmonic is 71.2 db down. Wtth 486 mv RMS, it is 61 db down, and 53.8 db down with 1 volt RMS out.
But what if you want something a bit dirty, but short of hard limiting? This attachment (
) shows the waveform at 1KHz for several levels of input signal with the symmetry control adjusted to a dc level of a few hundred mv. The distortion on the waveform is clearly visible well below final limiting. This is an extreme setting: at the other extreme with zero volts on the symmetry xontrol, the drain voltage (on the right FET) is about 13V. For the results shown in this figure it is about 2V. Applying a negative going input signal to the left FET causes the drain voltage (right FET) to move further down into the resistive region; the gain is very low, and the drain voltage does not change very much. When the input signal (left FET) goes positive, the drain voltage (right FET) increases, moving into a higher gain part of the curve and the waveform stretches out. This is not exactly what happens in a triode, of course, but it is similar, and it does generate both even and odd harmonics with emphasis on the lower harmonics. And of course, a lower setting on the asymmetry control gives a more subtle effect.
In this realization of the circuit, the asymmetry voltgage goes only to stages 2 and 3. The final stage just amplifies up the losses in the tone stack to make a good signal for the power amp, and the first stage can be thought of as a linear preamp where we want the cleanest grounding and the best possible SNR.
“Good stuff.... Pretty wild that solid state is so demonized, when its really the CIRCUIT, not using ss devices, that dictates the sound. Inspires me to wanna build a ss amp that could pass a blind fold test. This amp sure as hell does!!”
Solid state components have come a long way since that design; it should be possible to do much better now!
My further interpretation Lowell’s statement: instead of designing solid state circuits in the most convenient way, or minor variations of what everybody usually does, rather one should look and see how a tube amp does what it does and try to achieve a similar funcionality with solid state.
The goal is to design a solid amp equivqlent to a typical tube amp consisting of a high gain preap section and a power section. Another goal is to make the front end much quieter than a tube preamp. The main purpose of this is to allow the guitar volume to be used with good SNR over a wide dynamic range. (That requires some changes to the standard guitar electronics as well, of course.)
Consider the high gain preamp part first. (The power section is not done yet.) A high gain tube preamp uses several independent stages of gain with gain controls in between and a tone stack as well. A block diagram of one realization is given here: FETpreamp.pdf (along with the actual circuit, to be discussed below). So the solid state will do the same, but what to use? Some of requirements are:
1. The first stage must be extremely quiet. But given the flexibility in adjusting levels in the block diagram, really both the first and second stage should be very quiet. Well, in that case, it makes sense to make them all the same, and thus all quiet.
2. We want the device to be sort of tube like in its characteristics, if possibe, or better, have adjutable linearity and limiting characteristics.
So the first two requirements suggest junction FETs: pretty nice curves, and certain modern FETS are one nanovolt per root Hz, or even a bit better. Also the noise is really good noise: it is all voltage noise, almost no current noise, the kind that depends on the impedance connected at the input. This preamp must be good for a standard guitar pickup with cable, and that has a nasty high impedance resonance right where you do not want it from a noise perspective, exaggerating the effect of current noise.
So, if we are to use FETS, the rest of the requirements are concerned with how to use them.
3. The FET has higher capacitances than the tube, and so we must avoid amplifying it at the input (Miller effect).
The quietest circuit is probably a cascode, but
4. Given the square law response of the FET, its linearity is not as good as a triode, and so we need to improve the linearity with a good circuit.
For this application there is a better circuit than cascode (the usual choice to avoid the Miller effect): a source coupled pair. The circuit of this noninverting Stage of Gain (SoG) is shown in the same attachmnet. (The FETs are LSK170F, reasonably well matched for the left and right positions.) The noise voltage is worse by the square root of two since both devices contribute, but it is very linear for small signals. Maybe this circuit can be too linear, and so we want a way of adjusting the linearity as required.
The source coupled FET pair is a remarkable circuit. (Now would be a good time to review EB’s articles on JFETs: http://www.linearsystems.com/assets/...%20Bordely.pdf For a constant drain source voltage, the current varies as the square of the gate source voltage. That is, the transcondance varies linearly with VGS. This means that if two FETs are connnected as a source coupled pair and are matched, then the individual gate source voltages change oppositely, and the variation of the transcondances disappears. This is true even if the quiescent currents through the two FETs are not the same. This results in extremely good linearity, assuming that the FET is very close to a true current source. And it is, as long as VDS is large. However, the output resistance drops as VDS drops, resulting in another source of non-linearity. (See Figure 2A of the first EB paper linked to above.) So this suggests a way of making a preamp circuit with adjustable linearity: vary the biasing so that the quiesent VDS changes.
This is not an amplifier with a differential output. If we feed the signal into the gate of the left FET, its drain must go directly to the power supply (AC ground) in order to have no Miller effect. The output signal is taken from the drain of the right FET, with the usual resistor to the power supply.
The current limiting is excellent with this circuit. When the left FET turns off, the right FET gets all the current, doubling the equlibrium value. When the left FET turns on full, the right FET goes off, and the output is connected to the power supply through the drain resistor. (No power supply rejection in this circuit; we need good bypassing.) Some might remember ECL (emitter coupled logic). There is sort of an anlogy here to the way this circuit limits.
The power supply bypass capacitors are included in the circuit diagram of the SoG, and so this becomes a complete unit that can be duplicated for any of the four locations in the block diagram. In addition to the input and output a third line is brought out. This is the high quality ground reference, but it can also serves as a way to lower VDS (by introducing a small dc voltage), causing an asymmetrical non-linearity, producing even and odd harmonics.
Here are some measured characteristics, taken at the test output, that is, after three stages. First the noise; it is meausred by taking the spectrum with the input grounded, and then again with a calibration source, a 10K metal film resistor. This is not an easy measurement; the power spectra result from averaging many several second long FFTs. Then the calculations are performed to turn noise and noise plus cal into noise voltage. The spectra have power line harmonics and other spikes probably from nearby switching power supplies. They are too small to be heard, but they contaminate the delicate measurement. Thus the result is filtered in frequency with a low pass median filter which suppresses the outliers, giving the smooth measurement shown here:
Next some numbers on the linearity with the symmetry adjust on zero volts. With a suitable sine wave fed to the input, the two level controls before the test output are adjusted so that the signal increases from stage to stage. When the outpout at the test output is 300 mv RMS, the highest harmonic is 71.2 db down. Wtth 486 mv RMS, it is 61 db down, and 53.8 db down with 1 volt RMS out.
But what if you want something a bit dirty, but short of hard limiting? This attachment (
In this realization of the circuit, the asymmetry voltgage goes only to stages 2 and 3. The final stage just amplifies up the losses in the tone stack to make a good signal for the power amp, and the first stage can be thought of as a linear preamp where we want the cleanest grounding and the best possible SNR.
Comment