Some time ago, I was doing a design for an MM with no feedback,
using a one-stage transconductance. That worked just fine, so I
wanted to do the same for an MC pre-amplifier.
The problem is that all the design I've seen, (on MC head amps),
uses on or more large electrolytics, and I definitely wanted to
avoid that, so the challenge was to make an amplifier with
these basic features:
1)a gain of 80dB @ dc,
2)an input bias-current < 1muA
3)low noise
4)using easy to obtain components
The concept I was working from is as simple as it gets:
The dc-gain is defined as Ag = gm x Rl and with Rl = 33k5 in
the RIAA network and Ag = 80dB = 10,000x =>
gm = 10k/33k5 =0.298S (S = 1/Ohm).
The RIAA network is from Linsley-Hood as found on Elliot Sound
(sound-au.com/project25.htm), allthough I reduced the impedance-
level to 1/3, i.e. resistors devided by 3 and capacitors multiplied
be 3.
The design I came up with look like this (simplified):
The output of the LTPs (Long Tailed Pair) feeds into a combination
of a current mirror and folded cascode. With the R4,R5, and R6 as
specified, the quiscent currents of Q9, Q12 and Q6 are the same.
The same applies for Q1, Q4, and Q14.
The total gm from input to the RIAA network is :
gm = a x b c x Ic/Vt,
where
a = ½ as the input voltage is shared between each transistor
in the LTP,
b = 2 as there is two LTPs,
c = 3 well its just way the current mirror / folded cascode works,
Ic = the collector current in the LTPs, and
Vt = 26mV
Now we can calculate the Ic in the LTPs:
Ic = gm x Vt/3 = 0.298 x 26mV/3 = 2.58mA, and hence
R3 = R7 = (15-0.7)/(2Ic) = 2.77kOhm
Now, if you simulate (or build) the amp, the gain at LF/DC just can't
make it to 80dB. The problem is the Q6 and Q14 : the output impedance
is too low. This is dealt with in the full design.
A few words on the output buffer: The requirements of this buffer are:
1) A high input impedance, as not to load the RIAA network
2) Drive the output (!)
3) Low distortion.
1) states that Q17 and Q18 should run at a rather low Ic,
2) states that Q19 - Q22 should run at a "not too low" current.
3) The original diamond buffer had only four transistors, and
run at the same Ic, as this cancels out much distortion, albeit at
a rather high total current.
By adding two transistors, the output transistors can be run at almost
the same current as the input transistors, and still be able to drive
any "normal" amplifier down the line. Running at 560muA each this buffer
will drive a 5.8kOhm load to 13V and still stay in class A. It will
drive much lower load, but then in class B. It should be able to drive
a 825 Ohm load to 13V.
There is another issue: the ouput offset voltage. In the simplified
version, LTSpice calculates it to be app 1V. But this is with
identical transistors, and with physical transistors this isbound to
be much larger. After all, the DC gain is 10,000x so just a 1mv
difference in the input transistors, this will result in 10V at
output! This is dealt with in the full circuit.
The full circuit will follow in a subsequent post
Have a nice day!
using a one-stage transconductance. That worked just fine, so I
wanted to do the same for an MC pre-amplifier.
The problem is that all the design I've seen, (on MC head amps),
uses on or more large electrolytics, and I definitely wanted to
avoid that, so the challenge was to make an amplifier with
these basic features:
1)a gain of 80dB @ dc,
2)an input bias-current < 1muA
3)low noise
4)using easy to obtain components
The concept I was working from is as simple as it gets:
The dc-gain is defined as Ag = gm x Rl and with Rl = 33k5 in
the RIAA network and Ag = 80dB = 10,000x =>
gm = 10k/33k5 =0.298S (S = 1/Ohm).
The RIAA network is from Linsley-Hood as found on Elliot Sound
(sound-au.com/project25.htm), allthough I reduced the impedance-
level to 1/3, i.e. resistors devided by 3 and capacitors multiplied
be 3.
The design I came up with look like this (simplified):
The output of the LTPs (Long Tailed Pair) feeds into a combination
of a current mirror and folded cascode. With the R4,R5, and R6 as
specified, the quiscent currents of Q9, Q12 and Q6 are the same.
The same applies for Q1, Q4, and Q14.
The total gm from input to the RIAA network is :
gm = a x b c x Ic/Vt,
where
a = ½ as the input voltage is shared between each transistor
in the LTP,
b = 2 as there is two LTPs,
c = 3 well its just way the current mirror / folded cascode works,
Ic = the collector current in the LTPs, and
Vt = 26mV
Now we can calculate the Ic in the LTPs:
Ic = gm x Vt/3 = 0.298 x 26mV/3 = 2.58mA, and hence
R3 = R7 = (15-0.7)/(2Ic) = 2.77kOhm
Now, if you simulate (or build) the amp, the gain at LF/DC just can't
make it to 80dB. The problem is the Q6 and Q14 : the output impedance
is too low. This is dealt with in the full design.
A few words on the output buffer: The requirements of this buffer are:
1) A high input impedance, as not to load the RIAA network
2) Drive the output (!)
3) Low distortion.
1) states that Q17 and Q18 should run at a rather low Ic,
2) states that Q19 - Q22 should run at a "not too low" current.
3) The original diamond buffer had only four transistors, and
run at the same Ic, as this cancels out much distortion, albeit at
a rather high total current.
By adding two transistors, the output transistors can be run at almost
the same current as the input transistors, and still be able to drive
any "normal" amplifier down the line. Running at 560muA each this buffer
will drive a 5.8kOhm load to 13V and still stay in class A. It will
drive much lower load, but then in class B. It should be able to drive
a 825 Ohm load to 13V.
There is another issue: the ouput offset voltage. In the simplified
version, LTSpice calculates it to be app 1V. But this is with
identical transistors, and with physical transistors this isbound to
be much larger. After all, the DC gain is 10,000x so just a 1mv
difference in the input transistors, this will result in 10V at
output! This is dealt with in the full circuit.
The full circuit will follow in a subsequent post
Have a nice day!
Attachments
Last edited:
