Erno Borbely

JFETS: THE NEW FRONTIER, PART 1

tion, formed along the channel. This de-
Welcome to a new era in audio amplification where JFETs rule.
scription implies the primary difference
This noted designer champions their use to produce the best between a bipolar transistor and a JFET:
the pn junction in a JFET is reverse-bi-
sound in your audio amp circuits. ased, so the gate current is zero, whereas
the base of a bipolar transistor is for-
ward-biased, and the base conducts a

A
s most of our customers know, I breakthrough when I started to use base current. The JFET is therefore an in-
have been advocating the advan- mostly JFETs in the amps. It is my consid- herently high-input impedance device,
tages of FETs in general and JFETs ered opinion that it would be best to use and the bipolar transistor is comparative-
in particular, especially for low and only JFETs in all stages of the audio ly low-impedance.
medium level circuits. JFETs provide ex- chain. However, due to their limited Depending on the doping of the semi-
tremely high resolution, bringing out power-handling capability, it is practical- conductor material, you get so-called N-
more details, sounding cleaner, clearer, ly impossible to use them in output type or P-type material, and these result
and more natural than the best bipolar stages. Here, MOSFETs will rule for the in the N-channel or P-channel types of
transistors such as the LM394, and even foreseeable future. JFET. The symbol for an N-channel JFET
the best Telefunken tubes. Overall, I be- In spite of their quadratic characteris- is shown in Fig. 1A. The three “elec-
lieve the JFETs offer the best sound in tics and relatively high input capacitance, trodes” are called G, D, and S, for gate,
audio circuits. JFETs are fairly simple to use in audio am- drain, and source. The output character-
I have been working with JFETs since plifiers, and you, as an amateur, can de- istic for the N-channel JFET with the gate
the middle of the ’70s, when I developed sign most low-level stages in an audio shorted to source (i.e., VGS = 0) is shown
low-level amplifier modules with JFETs at chain yourself. Just like a single vacuum- in Fig. 1B.
Motorola. However, they were not com- tube triode or pentode, a single JFET can The characteristic field is divided into
petitive with the best bipolars at that handle the task of a line amp, and it is sig- two regions, first a “resistive” region
time. In the early ’80s came the first real- nificantly simpler to hook up. You can below the saturation voltage VSAT, where
ly low-noise, high-gm devices on the mar- also build a single-ended (SE) phono an increase in VDS results in a nearly lin-
ket. I have used these devices in the stage with only two JFETs. The rest is up ear increase in drain current ID. Above
input stages of practically all my designs to your imagination. Suffice it to say that
since then. However, I use bipolar tran- I hope the following introduction to
sistors in the second stages, mostly be- JFETs will whet your appetite for the
cause they offer a fairly simple design. “new frontiers” in audio amplification.
The output stages have always been
MOSFETs, because of the relatively high JFETs
current required in these stages. Field-effect transistors (FETs) have been
In the ever-continuing quest for better around for a long time; in fact, they were
sound, I have reviewed my designs regu- invented, at least theoretically, before
larly, improving the topology of the am- the bipolar transistors. The basic princi-
plifiers and also using better compo- ple of the FET has been known since J.E.
nents, thus bringing significant improve- Lilienfeld’s US patent in 1930, and Oscar
ments. However, I first achieved a real Heil described the possibility of control-
ling the resistance in a semiconducting
About the Author material with an electric field in a British
FIGURE 1A: Symbol for N-channel JFET.
Erno Borberly has been employed by National Semi- patent in 1935. Several other researchers
conductor Europe for the last 17 years. He was Manag- described similar mechanisms in the
er of Technical Training and worked as a consultant in
’40s and ’50s, but not until the ’60s did
human-resources development. He received an MSc
degree in electronic engineering from the Institute of the advances in semiconductor technol-
Technology, University of Norway in 1961, and worked ogy allow practical realization of these
seven years for the Norwegian Broadcasting Corp. de- devices.
signing professional audio equipment. He lived in the The junction field-effect transistor, or
US and was Director of Engineering for Dynaco and
The David Hafler Co. From 1973–1978, he worked for JFET, consists of a channel of semicon-
Motorola in Geneva, Switzerland, as Senior Applica- ducting material through which a cur-
tions Designer and Applications Manager. He has now rent flows. This channel acts as a resis-
taken an early retirement from National and is looking tor, and the current through it is con-
for OEM customers for whom he can design high-end
trolled by a voltage (electric field) ap-
audio equipment. FIGURE1B: Output characteristic for VGS=0V.
plied to its gate. The gate is a pn junc-

26 Audio Electronics 5/99

VSAT, an increase in VDS does not result
in a further increase in ID, and the char-
acteristic flattens out, indicating the “sat-
uration” region. Sometimes these two re-
gions are also called “triode” and “pen-
tode” regions.
You can use the JFET as a voltage-con-
trolled resistor or a low-level switch in
the triode region, and as an amplifier in
the pentode region. As you see, the N-
channel JFET conducts maximum cur-
rent IDSS with VGS = 0V. If you apply a
negative voltage to the gate, it reduces
the current in the channel, and you get a
family of output characteristics as shown A-1535-2b
in Fig. 2A. This device is called a “deple-
FIGURE 2A: Family of output characteris- FIGURE 2B: Transconductance curves for
tion” type of JFET. 2SK170.
tics for 2SK170.
In summar y, the JFET consists of a
channel of semiconducting material,
along which a current can flow, and this ferentiating ID with respect to VGS: temperature coefficient (tempco) be-
f low is controlled by two voltages, VDS comes zero. Obviously, if you need to de-
and VGS. When VDS is greater than VSAT, dI D 2I  VGS  sign for low drift, then the JFET must be
gm = = — DSS 1 — 
the current is controlled by VGS alone, dVGS VP  VP  operated at this point.
and because the VGS is applied to a re- You can calculate the zero tempco
verse-biased junction, the gate current is The transconductance gm becomes point with the following formula:
extremely small. In this respect, the N- −2IDSS/VP where the transconductance
channel JFET is analogous to a vacuum- curve meets the y-axis. This is the value VGS = VP + 0.63V
tube pentode and, like a pentode, can be you normally find given in the data
connected as an amplifier. sheets. Notice that there are five differ- Typical transconductance curves for two
The P-channel JFETs behave in a simi- ent transconductance curves given for different JFETs are shown in Figs. 3A and
lar manner, but with the direction of cur- the 2SK170 in Fig. 2B. This indicates 3B for a high-VP and a low-VP JFET, re-
rent flow and voltage polarities reversed. there is a range of ID curves for each spectively. It is obvious from the curves
The P-channel JFET has no good analogy JFET, due to manufacturing tolerances. that the zero tempco point occurs at a
among vacuum tubes. Also notice that the transconductance lower ID for high-VP JFETs and at a higher
curve stops where it meets the y-axis. ID for low-VP JFETs. If the VP is close to
The Transconductance Curve This is because the gate pn junction 0.6V, then the zero tempco point is close
As mentioned previously, you can use the would be forward-biased if VGS were to IDSS.
JFET as an amplifier in the pentode, or made positive for N-channel and negative
saturation, region. Here the VDS has little for P-channel JFETs, and gate current The Bias Point
effect on the output characteristics, and would flow. This is analogous to the con- As shown in Fig. 2B, the JFETs have a rel-
the gate voltage controls the channel cur- dition of vacuum tubes when the grid is atively wide range of transconductance
rent ID. Because of this, it is easy to char- made positive. Of course, a silicon pn curves. In order to operate the JFET as a
acterize the JFET in terms of the relation- junction does not conduct before the for- linear amplifier, you need to have a clear-
ship between ID and VGS, that is, with ward voltage reaches 0.6–0.7V, so you ly defined operating point. A typical
the transconductance curve. Figure 2B can apply several hundred mV in the for- common-source amplifier stage is shown
shows the transconductance curves for a ward direction without ill effects. JFETs in Fig. 4A. Assume that the +Vs is 36V,
typical low-noise, high-g m JFET, the are often operated with both polarities and you have selected a load resistor RL =
2SK170. of gate voltage—i.e., with gate current—in 10k. What happens now if you insert a
The drain current as a function of VGS RF applications. typical JFET, such as the 2SK170, for Q1?
is given by the formula: The change in the transconductance Figure 4B shows five of the transcon-
curve is not just a matter of tolerances ductance curves for the 2SK170, with
2
 V  due to manufacturing, but it also de- IDSS between 2.1mA and 10mA. If you
I D = I DSS 1 — GS  .
 VP  pends on the temperature, and this is take one of these at random and operate
due to two different effects. As the tem- it without RS, the actual drain current
VP is the gate pinch-off voltage, and is de- perature increases, the mobility of the will be the IDSS value. With 2.1mA, the
fined as the gate-source voltage that re- charge carriers in the channel decreases, voltage drop across RL will be 21V; i.e.,
duces ID to a very low value, such as which leads to an increasing channel re- the drain (OUT) will be sitting at 36 − 21
0.1µA. The formula indicates that the sistance, and hence a reduction in ID. = 15V. This might not be optimal from
transconductance curve has a square-law On the other hand, the barrier poten- the point of view of maximum output or
form. It also shows that if you know IDSS tial of the gate pn junction decreases minimum THD, but it will work all right.
and VP, you can draw the transconduc- about 2.2mV/°C, which causes the ID to However, with IDSS = 10mA, the volt-
tance curve for any JFET. The transcon- increase. There is a point on the age drop should be 100V, which is clear-
ductance gm, which is the slope of the transconductance cur ve where these ly impossible with Vs = 36V, and the
transconductance curve, is found by dif- two effects cancel one another, and the to page 30

Audio Electronics 5/99 27

 The best solution, of course, is to se- ideal devices are hard to come by, and
lect the devices for your particular appli- practical devices also generate some
cation. Assume you wish to build a sin- higher harmonics. Again, in this respect
gle-ended phono amp with JFETs and a there is a close similarity to vacuum
passive RIA A correction network, and tubes. Looking at the transconductance
you decide to use the 2SK170 devices. In curve, you can easily see that it is more
order to keep circuit noise to a mini- linear close to the y-axis than further
mum, you would use the 2SK170 with- down on the curve. From the point of
out RS, i.e., at IDSS. Furthermore, you view of linearity, it is therefore an advan-
would need a relatively high current to tage to operate the JFET with a higher ID.
be able to drive a passive RIA A correc- Figures 6A and 6B show the transcon-
tion network. If you choose, say, 5mA, ductance characteristics for two JFETs I
A-1535-3a you would need to select the devices use in many of my ampli fiers. The
FIGURE 3A: The zero tempco point for from the “GR” group. But how? The se- 2SK170 is a high-transconductance de-
2SK246. lection is easy. vice with low VP, and the 2SK246 is a
low-transconductance JFET with a high-
Testing JFETs er VP.
Figure 5 shows a simple circuit with I have selected a 2SK170 with IDSS =
which you can select JFETs and also 6.2mA and a 2SK246 with IDSS = 5.6mA
match them if necessary. The tester feeds
current into the source or connects the
source to ground to measure the essen-
tial parameters of the device. In position
1 (switch in counterclockwise position),
the source is connected to −10V through
a 1M resistor. This feeds the source with
approximately a 10µA current, which
you can consider the cutoff point VP for
the JFET. (Data sheets specify lower val-
ues, but this gives you a more practical
value for measurements.) The voltmeter
FIGURE 3B: The zero tempco point for now indicates the pinch-off voltage VP
2SK170. for the device. FIGURE 4A: A common-source amplifier.
The next two positions measure the
from page 27 VGS for the device at given drain cur-
rents. These positions give practical read-
amplifier goes into saturation. Obviously, ings for design purposes, and you can
if you wish to use any or all of these JFETs, choose the constant-current sources for
you must reduce the effect of the wide the values you need. The push-button
range of IDSS. switch shorts the source to ground, and
The solution is to use a source resistor the mA meter measures IDSS. If you wish
RS, similar to the biasing arrangements to measure only VP and IDSS, you can per-
used in bipolar transistors or tube ampli- manently wire the source to −10V
fiers. To illustrate the effect, I have through the 1M resistor, which gives you
drawn in the line for a 100Ω resistor in VP, and then short the source to ground
the transconductance characteristics. with the push-button to read IDSS.
The range of drain currents is now limit- If you test P-channel devices, you
ed between 1mA for the IDSS = 2.1mA de- must reverse the supply voltages and the FIGURE 4B: The source resistor Rs stabi-
lizes the bias point.
vice, and about 2.6mA for the I DSS = constant-current diodes. Normal-
10mA device. The drain voltages will be ly, I test a large batch of devices
36 − 10 = 26V and 36 − 26 = 10V, respec- (say 100 of each type) and sort
tively. This is still too much variation them by IDSS. The different de-
from the point of view of THD and maxi- vices are then used in different
mum output swing, but at least there is applications.
no saturation with any of these devices.
Fortunately, JFETs are sold with much Some Practical
narrower IDSS ranges, which makes life Measurements
easier in terms of proper biasing. The As mentioned previously, the
2SK170 comes in three IDSS groups: the transconductance cur ve has a
“GR” group is 2.6–6.5mA, the “BL” quadratic form, and if you wish
group is 6–12mA, and the “V” group is to use it to amplify audio signals,
10–20mA. If you use a “GR” device with it will create harmonics. A true
RS = 100Ω, the ID will vary between 1 quadratic curve would generate FIGURE 5: A simple test jig for N-channel JFETs.
and 2mA, which is almost acceptable. only second harmonics; however,

28 Audio Electronics 5/99

to illustrate the difference of operation 20.68/0.2 = 103.4, which is 40dB. The two amplifiers with K170 and K246. The
with very similar values of IDSS. The gate output range for the K246 is 2.5mA to K170GR had an IDSS of 5.5mA, and I op-
pinch-off voltage is approximately 0.45V 5.6mA. With the same drain resistor of erated it first with RS = 0 and RL = 3.3k.
for the K170, and 2.75V for the K246. In 4.7k, the output-voltage swing will be This gave me a gain of 36.4dB and a fre-
order to operate them at the most linear 26.32 − 11.75 = 14.57V pk-pk. The gain quency response of over 400kHz. The
part of the characteristic, I selected bias is 14.57/1 = 14.57 ti mes, wh ich is THD is shown in Column 1 of Table 1.
points at VGS = 0.1V and ID = 3.8mA for 23.38dB. That is, the high-VP device has Column 2 shows the same K170GR
the K170, and VGS = 0.5V and ID = 4mA lower gain than the low-VP one. device, but this time with RS = 50Ω. This
for the K246. These points are set with RS reduces the drain current to approxi-
= 27Ω and 125Ω, respectively. When Higher Is Lower mately 2.5mA, so I increased the drain
The most obvious difference between Of course, this can be explained by the resistor to 8.2k to have the same DC con-
the two JFETs is in the maximum input transconductance. The gm for the K170 ditions as before. The THD is reduced by
swing with which you can drive them. is 2IDSS/VP = 27.55mS. The gain is gm × roughly 6dB. Column 3 shows the
The K170 allows approximately ±0.1V RL, which gives 127 times, a bit higher K246BL amp operating at ID = 5.1mA,
peak before the gate goes positive, but than the graphical analysis. The explana- with RS = 100Ω, and RL = 4.7k. The out-
the K246 has a range of ±0.5V! Naturally, tion for this is that this gm is at the point put is now a bit lower than half of the
I could move the working point further where the cur ve crosses the y-axis, supply voltage, and the maximum output
down on the transconductance curve in which is always higher than at the work- is therefore limited. But the THD is quite
order to increase the input range, but ing point, and that the curve is not a low, again about 6dB lower than the pre-
vious circuit.
The K170GR circuit seems to be popu-
lar for phono input stages, and a number
of these are circulating on the Internet.
RS is usually shorted to achieve minimum
noise. However, even without RS, the
noise of a single K170 is not low enough
for MC pickups. To achieve lower noise,
you can parallel several of these devices.
Doubling the JFETs with comparable gm
reduces the noise by approximately 3dB.
I hooked up four K170s in parallel to see
how it works (Fig. 8). Each device had an
IDSS of approximately 15mA, and the
drain currents with RS = 6R8 are 10mA
A-1535-6a each. With an RL = 511Ω, the drain is sit-
A-1535-6b
ting at 14.8V DC.
FIGURES 6A/6B: Input/output range for 2SK170 and 2SK246, respectively.
The gain is 34dB and the frequency re-
sponse is 360kHz. The THD for this cir-
cuit is shown in Column 4 of Table 1. Re-
member that this circuit is working at
very low levels, where THD is indeed
low. The equivalent input noise is also
reasonably low at approximately 100nV
over a 20kHz bandwidth. Not bad for a
simple circuit. Want to try it?

Input Capacitance
As mentioned before, the JFETs have a
relatively high input capacitance, which
can be an important design factor. Just
FIGURES 7A/7B: Practical amplifiers with 2SK170 and 2SK246. like tubes and bipolar transistors, JFETs
also have interelectrode capacitances
eventually I would reach the other limit- straight line, making the output swing that affect the frequency response of the
ing point, where the gate cuts off at VP. smaller than the theoretical value. JFET when it is used as an amplifier. The
The thing to understand here is that a In any case, this quick calculation two capacitances, which are of impor-
high-VP JFET has a wider range of input gives you a reasonable starting point tance for audio use, are the Ciss and Crss.
swing than one with a low VP. from which to design the circuit. The The Ciss is called the input capaci-
Other obvious differences involve the corresponding gm for the K246 is 4mS, tance and Crss the reverse transfer ca-
output range and the gain. With a ±0.1V so obviously the gain is also much small- pacitance. Typical values for the Ciss are
gate voltage, the drain current varies be- er at 19.14, that is, 25.63dB. Again, this 30pF for the K170, and 9pF for the
tween 1.8 and 6.2mA for the K170. With results in a higher value than the graphi- K246. The high-gm devices have a much
a drain resistor RL = 4.7k, this results in cal analysis. higher input capacitance than the low-
an output swing of 29.14V − 8.46V = Now for some real circuits and THD gm ones. The Crss is 6pF and 2.5pF, re-
20.68V pk-pk. The gain will then be measurements. Figures 7A and 7B show spectively. The Crss seems to be relative-

Audio Electronics 5/99 29

ly low, but this is the one that dominates duced to 50pF. With such low input
the input capacitance of an amplifier capacitance there is no longer any
through the Miller-effect. danger of creating a low-pass filter
The input capacitance of a normal with the volume control.
common-source JFET stage as shown in As though the existence and size
Fig. 7, but with RS = 0, is given by the of the input capacitance were not
formula: Cin = Ciss − AV × Crss, where AV enough, it is also voltage depen-
is the voltage gain of the stage. Note that dent, which might cause distortion
a common-source stage inver ts the in certain applications. Figures 10A
phase, so AV is negative, making Cin a and 10B show the voltage depen-
positive number. Since AV can be a sig- dence of Ciss and Crss, respectively, FIGURE 8: Paralleling JFETs reduces the noise.
nificantly large number, the input capac- of the K170 JFET.
itance of the stage can be very high. Depending on the excursion of
TABLE 1
I have measured the input capaci- the input/output signal, you get a
tance for the amplifier in Fig. 7, both capacitance modulation, and this Output Column 1 Column 2 Column 3 Column 4
voltage, K170GR, K170GR, K246BL, 4xK170V,
with and without RS. Without RS, the ca- can cause distortion of the audio V RMS RS = 0, RS = 50, RS = 100, RS = 6R8
pacitance was over 600pF! With RS = signal. This shows up mostly when RL = 3.3k RL = 8.2k RL = 4.7k RL = 511R
100Ω, the input capacitance dropped to you drive the circuit from a high- 0.1V 0.095% 0.06% 0.02% 0.04%
127pF, because of the local feedback source impedance. I have tested the 0.3V 0.2% 0.1% 0.047% 0.1%
through RS. To appreciate the signifi- circuit described in Column 1 and 1V 0.6% 0.32% 0.15% 0.32%
cance of this, assume that you are dri- Column 2 of Table 1 with different 2V 1.3% 0.65% 0.29% 0.67%
3V 1.9% 0.98% 0.4% 1%
ving the amplifier from a 100kΩ volume source impedances, and could not
5V 3.2% 1.7% 1.65%
control. The amplifier will see a maxi- measure any significant increase in 10V 6% 3.4% 3.5%
mum “source impedance” of 25k when THD up to 50k source.
the volume control is in the middle. If However, when the noncascode
you calculate the 3dB point of the low- circuit was driven from 500k, the THD volume control of no more than 50k. (Of
pass filter formed by the volume control increased approximately 6dB. The cas- course, you would probably use no more
and the input capacitance of 600pF, you coded circuit showed no significant in- than 50k any way, because of the in-
find that it is about 10kHz! If you use the crease at any source impedance up to creased noise with higher impedances.)
K170 without RS, you certainly must use 500k. To avoid capacitance modulation Note that in these circuits only two
a volume control, which is less than problems, I recommend that you use a types of JFETs have been involved,
100k.

Cascode to the Rescue

There is another way of reducing the
input capacitance of the amplifier. Cas-
code connection of devices was invented
in the tube era, but has also been used
extensively with bipolar transistors. One
of the advantages of cascoding, if you re-
call, is reduction of input capacitance,
which makes it easier to design high-fre-
quency amplifiers.
I have connected two circuits to test
this (Fig. 9). The upper JFET needs a bias
voltage, and it is easy to get this by con-
necting its gate to the source of the lower FIGURES 9A/9B: Cascoding and local feedback with Rs reduces the input capacitance.
JFET. (Of course, you can also
generate this bias from the sup-
ply voltage with a voltage di-
vider, as you normally do with
tube cascodes.) I am using a
high-VP JFET for the upper de-
vice, so that the lower JFET has
enough voltage across it to oper-
ate in the saturation region.
The input capacitance of the
circuit in Fig. 9A is approximate-
ly 160pF, so the cascoding in-
deed reduces the input capaci-
tance. Fur ther reduction is
achieved by adding local feed-
FIGURE 10: Voltage dependence of Ciss for FIGURE 11: Voltage dependence of Crss for
back with R S (Fig. 9B). The 2SK170. 2SK170.
input capacitance is now re-

30 Audio Electronics 5/99

whereas there are thousands of them on
the market. Also, I have used them for il-
lustration purposes only, and, although
they work as described, I have made no
attempt to optimize them for any partic-
ular application.
In Part 2 of this article, I will discuss
the differential topologies. If you have
questions, please don’t hesitate to send
me an e-mail or a fax (Borbely Audio, e-
mail: borbelyaudio@t-online.de, FA X:
+49/8232/903618, Web site: http://
home.earthlink.net/~borbelyaudio).
And, of course, if you wish to buy some
JFETs to experiment with, we have tons
of them in stock. For a little extra, we
even do a selection for you. Have fun ex-
periencing the “new frontier” in audio
amplification. ■

Acknowledgements
My sincere thanks to Walt Jung of Analog Devices, who
kindly read the manuscript and provided valuable com-
ments and suggestions.
Also thanks to our customers: Dr. Juergen Saile, Ger-
many, Reza Habibi of Electro Concept Services,
France, and Winfried Ebeling of Crystal Audio Re-
search, Germany, for their valuable feedback, com-
ments and suggestions throughout the ALL-FET devel-
opment program.

Audio Electronics 5/99 31

Erno Borbely

JFETS: THE NEW FRONTIERS, PART 2

JFET 2, operating it with source input.
This noted design expert from Germany continues his series by
Seen from gate 2, the same thing hap-
showing how you can best take advantage of JFET performance. pens—JFET 2 is in common-source mode,
driving JFET 1 in the source. There are a
number of advantages to operating two
JFETs in this way, and I will start here
with the common-mode rejection.

I
n Part 1 of this article, I discussed the The differential gain of the stage is:
single-stage (or single-ended) amplifi- Common-Mode Signals
er operating in common-source AV(DD) = (VD1 − VD2)/(VGS1 − VGS2) A very important feature of the differen-
mode. As these stages are usually limited = RD × gm, tial amplifier is its ability to reject com-
in audio to AC signals, the inherent DC mon-mode signals. Common mode
drift is of relatively little importance. You which is the same as the gain of a single means that both gates are driven with the
can even use them for DC signals if you common-source stage. For R0 to be very same polarity and equal amplitude sig-
select the working point carefully at zero large, −VS must also be very large. This is nals. It is easy to see that if only gate 1 is
temperature coefficient. However, if you usually inconvenient, so instead of a re- driven positive, then ID1 increases and
remember the formula for zero tempco sistor, you use a so-called constant-cur- ID2 decreases. But if both gates are driven
(VGS = VP + 0.63V), you realize that the rent source, which delivers I0 indepen- positive, then both ID1 and ID2 must in-
condition is different from unit to unit, dent of −VS (Fig. 12b). crease, which is impossible because ID1 +
since VP is different. Due to its symmetrical nature, you can ID2 = I0; i.e., I0 is constant. Consequently,
A better solution is to use a differential also consider the differential amplifier as the differential amplifier cannot amplify
ampli fier, where the dri f ts of t wo two symmetrically arranged “half-cir- same-polarity or common-mode signals.
matched JFETs tend to cancel each other. cuits,” each with a JFET, a load resistor, Just how good it is in rejecting com-
The configuration is shown in Fig. 12a. and half of a current source, providing mon-mode signals is expressed with the
If R0 is large enough, then: I0/2.1 This is shown in Fig. 13. If the two common-mode gain:
JFETs are “identical,” then you can join
ID1 + ID2 = I0. the two half-circuits together at the AV(CM) = −RD/2r0,
sources without upsetting the DC opera-
Further, if ID1 changes ∆ID1, then ID2 tion. However, you now have balanced where r0 is the output impedance of the
also changes by the same amount, but in single-ended amplifiers.2 constant-current source. In order to have
the opposite direction, i.e., Seen from gate 1, JFET 1 operates as a low common-mode gain (i.e., good rejec-
common-source amplifier, except that tion), the output impedance of the cur-
∆ID1 = −∆ID2. the source is connected to the source of rent source must be very large.

FIGURE 12A: Basic differential amplifier FIGURE 12B: Improved differential amplifier FIGURE 13: The differential amplifier repre-
with JFETs. with constant-current source. sented with two symmetrically arranged
“half-circuits.”

16 Audio Electronics 6/99

 The importance of low common-mode
gain is closely related to the temperature
drift, because changes in ID, VGS, and gm
can be considered common-mode signals
if they are the same for both JFETs. Nor-
mally, a common-mode rejection ratio
(CMRR) is specified for the differential
amplifier. It is the ratio between the dif-
ferential gain and the common-mode
gain:

CMRR = AV(DD)/AV(CM) ≈ 2gm × r0.

Obviously the two JFETs must be

closely matched to achieve good com-
mon-mode rejection. In fact, these two
FIGURE 14A: A practical differential amplifi- FIGURE 14B: The cascode connection
formulas are valid only if the two JFETs er with the 2SK389V dual monolithic JFET. reduces the input capacitance.
are perfectly matched. Although it is pos-
sible to select well-matched JFETs, the
easier way is to use dual JFETs matched anced mode. Gain without local feed- anced signal from the two drains.
by the manufacturer, or even better, dual back (RS1 = RS2 = 0) is about 64 times, I have made some rudimentary THD
JFETs manufactured on the same silicon which is 36dB. Frequency response is measurements in balanced mode,
chip, i.e., monolithic duals. 175kHz, and the input capacitance is shown in column 3 of Table 2. Unfortu-
I have been using the NPD 5566 dual 330pF. THD, measured at 1kHz, is nately, my oscillator and THD analyzer
N-channel and the A H 5020CJ dual P- shown in column 1 of Table 2. (HP339A) are unbalanced, so I needed
channel JFETs.3 However, these are not Next I inserted source resistors RS1 = to improvise the balanced operation
truly complementary types, as I pointed RS2 = 100R, and reran the measurements. with op amps, limiting the measure-
out in my article.3 The first complemen- Due to the local feedback, the gain ments to the levels shown in the table
tar y types on the market were the dropped to approximately 28 times, and (lower levels were masked by noise).
2SK240/2SJ74 medium g m and the the input capacitance to 160pF. The Nevertheless, it clearly indicates that the
2SK146/2SJ73 high gm/low-noise types. THD also decreased by about 6dB. circuit thrives in balanced mode, having
These are closely matched single de- In order to reduce the input capaci- 10–20dB less THD compared to unbal-
vices, mounted in a common aluminum tance further, I put 2SK246 cascodes in anced mode. It also indicates the advan-
case for good thermal tracking. Unfortu- the circuit (Fig. 14b). The gain did not tages of this circuit relative to the SE cir-
nately, these devices are no longer in
production. TABLE 2
Output in Column 1 Column 2 Column 3
Dual Monolithic JFETs V RMS SE mode, RS1 = RS2 = 0 SE/cascode Balanced mode,
Although there are plenty of N-channel mode, RS1 = RS2 = 100R RS1 = RS2 = 100R
dual JFETs on the market, complemen- 0.3 0.013% 0.006% (noise)
tary dual monolithic JFETs are rare. In 1 0.035% 0.012%
fact, I k now of only one family, the 3 0.27% 0.1% 0.023%
2SK389/2SJ109, made by Toshiba. These 5 0.85% 0.33% 0.06%
are still manufactured and available, so I 8 2.6% 1% 0.12%
10 4.7% 2.3% 0.17%
use them in all my amps with differential
input. Now I’ll describe some practical
differential circuits. change significantly, but the input capac- cuits discussed in Part 1.
Figure 14a shows a simple differential itance dropped to 50pF! THD also de- The SE purists might naturally say that
amplifier with the 2SK389 dual mono- creased, as shown in column 2, Table 2. this is due to the cancellation of even-
lithic JFET from IDSS group V. I hooked it order harmonics in the balanced circuits,
up with ±36V to operate the JFETs under Balanced Mode which is true. But as Nelson Pass points
conditions similar to those of the SE A couple of comments are in order con- out on his homepage, in comparison to
ones. The constant-current source is a cerning this circuit. According to the the SE stages, the balanced circuit does
J511 JFET delivering 4.7mA. In order to measurements, it is a very decent de- not give rise to odd-order distortion.
run the drains at roughly one-half the sign, considering that it uses only a very There is simply not much distortion left
supply voltage (about 18V), I chose RD1 small amount of local feedback. The gain in the balanced circuit.
= RD2 =10k. is still fairly high, and you can reduce it As mentioned in Part 1, the input
First I tested the amplifier in single- further by increasing the source resis- capacitance is voltage-dependent, which
ended mode, i.e., gate 2 connected to tors, which in turn further reduces the can cause THD when the amplifier is
ground, with the measurements taken at THD. However, to fully take advantage driven from high source impedances. I
VD2. (VD2 has the same phase as VGS1.) of the symmetrical nature of this circuit, have tested the circuit described in col-
Although from the operational point of you should use it in balanced mode, umn 2 of Table 2 with 50k, 100k, and
view it is single-ended, I think this mode which requires applying a balanced sig- 500k sources. There was no measurable
is more appropriately called the unbal- nal at the two gates and taking the bal- change in distortion up to 100k, but at

Audio Electronics 6/99 17

 FIGURE 15A/B/C: Basic JFET source-follower circuits.

500k, I could see a slight increase. TABLE 3

Again, for noise reasons, you should
Output V Column 1 Column 2 Column 3 Column 4
probably keep the source impedance RMS RS = 5.11k RS = constant- RS = JFET The “Borbely”
below 50k, so there is no problem with current source current source source follower
the capacitance modulation any way. I 0.3V 0.0025 0.0023 0.002 0.0025
also checked the CMRR by connecting 1V 0.0033 0.0024 0.0018 0.0035
the two gates together and driving them 3V 0.011 0.0025 0.0016 0.0045
with a 3V RMS signal. The output, again 5V 0.02 0.003 0.0016 0.0074
in balanced mode, was down 87dB at
1kHz. The CMRR dropped to 70dB at
10kHz and 63.5dB at 20kHz, but even at
100kHz, it was 50dB!

The Output
I have now described two types of am-
plifier stages using JFETs, the common-
source or single-ended stage, and the dif-
ferential or balanced amplifier. You can
use either of these to build audio ampli-
fiers, depending on your preference for
balanced or unbalanced operation. Per-
sonally, I prefer the differential circuit,
because you can use it with balanced or
unbalanced sources, and it can also feed
balanced or unbalanced power ampli-
fiers. Balanced operation gives a subjec- FIGURE 16A/B: These source-follower circuits can drive low-impedance loads with very
low THD.
tive impression of increased dynamics.
It can also be an extremely useful inter-
facing consideration in breaking up
ground loops.4
There are two issues to consider when
talking about the SE and balanced ampli-
fiers. First of all, the output does not sit
at 0V DC, but at some 10–20V above
ground. If you wish to connect it to, say,
a DC-coupled power amplifier, you must
block this DC voltage from reaching the
power-amp input. This is easily done
using a capacitor, and this problem is
well known to all SE fans, whether of
tube or semiconductor variety. I will
therefore not spend much time on the
subject.
A much more important question is
whether these circuits can drive the
input impedance of a power amplifier. FIGURE 17: The all-JFET balanced SE line amp.
The output impedance of the amps ex-

18 Audio Electronics 6/99

amined is basically equal to the drain re- very low THD (see column 3 of Table 3). 18Ω, and the input capacitance is 28pF.
sistor of the amplifier. If RD = 10k, then Most important, the circuit can drive
the output impedance is also close to Follower Feature low-impedance loads without distress—
10k. But if the input impedance of the One of the most important features of a the 1k/3V RMS THD was 0.0078%. With
power amp is also 10k, then you are cer- follower is its ability to drive low-imped- a 10k load, there is no difference from
tainly in trouble. First, you lose 6dB of ance loads. I checked all three circuits the no-load results.
gain by the voltage division between the with 1k and 10k loads at 3V RMS output. I also tested the source followers for
10k resistors; second, the 10k input will With 1k they measured 1%, 1.7%, and THD caused by the voltage-dependence
most likely load the output and cause a 0.23%, respectively. Although its output of the input capacitance. Since the volt-
lot of THD. Even with 20k or 50k input impedance is actually higher than the age excursion is much larger at the input
impedance, you might run into prob- circuits in Figs. 15a and b, that in 15c is because of the unity gain, the circuits are
lems. It is advisable to put an impedance better in driving low-impedance loads. also more susceptible to the distortion.
transformer at the output to avoid this. With a 10k load, the circuits in 15b and There is no significant increase up to a
Source followers to the rescue! 15c didn’t have many problems. THD 10k source; however, at 50k the THD is
was 0.004 and 0.0022%. increasing by an order of magnitude.
JFETs as Followers My choices of source followers are Normally this is no problem, because the
Just like tubes and bipolar transistors, shown in Fig. 16. The circuit in 16a is a source impedance is usually very low.
JFETs can also be operated as followers, JFET version of the tube White cathode However, in certain applications such as
more specifically source followers. The follower.5 Basically, the circuit is an ex- filters, this can cause distortion.
basic circuit is shown in Fig. 15a. The tension of Fig. 15c, in that the follower is Of course, using any of these source-
drain is AC-grounded, and the output fed with a constant-current source, but follower buffer circuits with the SE and
signal is taken out across the source re- in addition the drain current of the cur- differential amplifiers discussed previ-
sistor, which means it operates with rent source is modulated by the AC sig- ously solves only one of the problems
100% local feedback. The gain of the nal. When the output signal goes posi- stated at the start of this section—the
source follower is: tive, the tail current decreases, and when drive-capability problem. The DC volt-
it goes negative, the current increases. age is still there. Given the topology of
AV = gm × RS/(1 + gm × RS) The result is a significant reduction of these circuits, you must use a capacitor
the output impedance and an apparent at the output to block the DC voltage.
Two things become obvious from the increase in drive capability. Naturally, you can also solve this prob-
formula: first, the source follower does The output impedance with the de- lem by using level-shifting circuits, but it
not reverse the phase of the signal, and vices shown measured 2.3Ω. The input requires a bit more circuit design. For
second, if gm × RS >> 1, then the gain be- capacitance is about 5pF, the same as now, I’ll look at an all-JFET balanced/SE
comes approximately unity. In order to the previous source-follower circuits. all-FET line amp, using the circuits al-
make RS large, you can use a constant-cur- The penalty for the drive capability is a ready developed.
rent source with high output impedance slight increase of distortion (see column
(Fig. 15b). The linearity is also dependent 4 in Table 3). The THD with a 1k load The Balanced/SE All-JFET Line Amp
on RS (see the 1kHz THD measurements and 3V RMS is 0.0095%. The necessary The schematic shown in Fig. 17 consists
in columns 1 and 2 of Table 3). gate drive voltage is derived from a small of the differential amplifier Q1/Q2, cas-
The input capacitance is low because resistor in the source follower’s drain coded with Q3/Q4, and the output
it is not augmented by the Miller effect. I circuit, and it is AC-coupled to the cur- buffers Q5/Q6 and Q7/Q8. The differen-
measured approximately 5pF for the cir- rent source. tial amplifier uses a dual monolithic
cuits in Fig. 15a and b. The output im- K389V JFET. Each JFET operates at just
pedance equals approximately 1/g m. Power Dissipation over 2mA, this current supplied by the
With high-gm devices, this will be fairly John Curl used the complementary JFET J511 constant-current source. The 330R
low. I measured 38Ω for the basic cir- source follower shown in Fig. 16b in the source resistors provide local feedback
cuits in Fig. 15a and b. JC-2 phono-preamp module.6 The JFETs and control the gain of the differential
The circuits in 15a and b have a DC work in Class A as long as the peak load amplifier. The trimpot P1 cancels out
offset voltage at the output—the gate- current is less than twice the bias cur- small imbalances bet ween the t wo
source voltage at the given drain current. rent. A fter that, the circuit works in JFETs, but it is normally unnecessar y
For the JFETs I used in the test setup, I Class AB. I usually let the two matched with monolithic duals, and you can leave
measured a 0.2V offset. If you need zero devices work at I DSS , to maintain as it out.
DC output, you can use the circuit in Fig. much Class A headroom as possible. The Cascode FETs are K246BLs. The
15c. Here the constant-current source is However, you must watch the power dis- output buffers are those I developed
made with the same type of JFET as the sipation. If IDSS is such that the power from the tube White Cathode Follower,
follower. dissipation is more than the maximum al- shown in Fig. 16a (“modestly” called the
If the two JFETs are matched and the lowed, then you need to insert a source “Borbely” source followers here). The
two source resistors are equal, then the resistor to reduce the drain current or se- supply voltage is ±36V. Of course, you
DC offset will be very small. With two lect a device with lower IDSS. can make the negative supply much less
matched K170BLs, I measured less than I tested the circuit with K170/J74, than 36V; the constant-current source re-
1mV offset. (It would probably be even both the BL and V types, and got excel- quires only a couple of volts for proper
lower if you used here a dual monolithic lent results. The THD is shown in col- operation. I made them both 36V to be
JFET like the K389BL/V.) DC drifts tend umn 4 of Table 3. Depending on the able to try other configurations.
to cancel out as well, because of the matching, the offset can be as low as The output caps must be of highest
matched devices. The circuit also has 1mV. The output impedance is about quality in order to preserve the outstand-

Audio Electronics 6/99 19

ing sound quality of this simple circuit. If
you are likely to drive loads down to 1k,
then the caps must be a minimum of
10µF. If you are driving normal 10k or
higher loads, you can get away with a
1µF or 2.2µF cap. I tried the Hovland Mu-
sicaps, which are rather neutral, but
there are plenty of good caps on the mar-
ket you can try.
Normal oil caps are not for this circuit;
they destroy the excellent resolution to a
“nice” blurred mish-mash. (Don’t get the
idea that I don’t like oil caps; I use them
in some of my amps.) I would have liked
to try some silver-foil caps, but, alas, the
prices are more ridiculous than the cable
prices, and I refuse to play that game. (If
anyone knows of a reasonably priced sil-
ver-foil cap, please let me know.)
You can use the line amp with unbal-
anced or balanced sources, and you can
feed power amps with balanced or un-
balanced inputs. However, you should re-
ally take advantage of its superior perfor-
mance in balanced operation, as I men-
tioned before. Should you use it with un-
balanced sources, then you must short
the –INP to ground. And in the unlikely
event that you don’t wish to take advan-
tage of the balanced outputs, you can
leave out the circuit around Q5/Q6, i.e.,
the negative output. I recommend a 10k
or 20k ladder attenuator as a volume con-
trol. Good luck with the JFETs, the “New
Frontiers” in audio amplification. ■

References
1. Erno Borbely, “New Devices for Audio from National
Semiconductor,” Audio Amateur 5/83, p. 7.
2. Nelson Pass, http://www.passlabs.com.
3. Erno Borbely, “Third Generation MOSFETs: The DC-
100,” The Audio Amateur 2/84, p. 13.
4. Walt Jung, private communication.
5. Erno Borbely, “Differential Line Amp with Tubes,” GA
1/97, p. 24.
6. John Curl, “Classic Circuitry,” The Audio Amateur
1977, Issue 3, p. 48.

Acknowledgments
Many thanks to Walt Jung of Analog Devices for his
valuable comments and suggestions.

20 Audio Electronics 6/99