Non-Feedback Amplifier


This audio amplifier is a non-feedback design, that is, there are no feedback loops. Further, it is divided into two parts: a voltage amplifier and a buffer amplifier. The voltage amplifier may be compared to an audio preamplifier, but with higher gain. However, the voltage amplifier can be used as an ordinary preamplifier, if you like. The buffer amplifier in fact is a current amplifier, with no voltage gain at all. You could possibly drive it from an ordinary audio preamplifier, but I think the voltage gain in most cases would be too low. You may also use the voltage amplifier as a headphone amplifier. Since the gain is set by one resistor, both a preamplifier and a headphone amplifier can have the gain that suits you.

The compact disk (CD) players with peak output signals typically in the region of 2V (a value significantly greater than that offered by many ordinary preamplifiers), in reality has made traditional preamplifiers obsolete with CD players. Some of my power amplifiers designs, thanks to their high gain and high input impedance (with negligible offset current and voltage), have rendered the possibility of a "passive preamplifier" as a more optimum interface between the signal sources (as the CD player) and the power amplifier.

This approach is, in principle, most welcome as the reduction of redundant circuitry only enhances overall system quality. Since the traditional preamplifier in audio reproduction equipment no longer offers a useful function, it is reasonable to redefine its role and use a solution that is more optimum. This can be done in the following way:

Efficiency targets generally dictate that the power transistors operate in class AB; consequently the currents drawn by the output and driver transistors are a very distorted version of the input signal. The bandwidth of these currents is therefore much in excess of the audio band. Clearly, any induction to high sensitivity circuitry, such as this non-linear modulation of supplies, will cause an error signal to be embedded within the primary signal. Such error will be more problematic under complex transient excitation with difficult loudspeakers (e.g. with problematic impedance characteristics).

It is well known that the principle advantage of class A, is that the currents within the amplifier are now a more linear function of the input signal, without rapid switching edges. Thus, we see primarily linear error rather than transient non-linear error, generated within class AB. Consequently, for a class AB output amplifier, the supply must be isolated from the more sensitive voltage amplifier, and it is necessary to minimise ground rail contamination and magnetic induction. Also, for an optimum interface, the input impedance of the power buffer should be high, constant and not represent a reflection of the loudspeaker or be distorted by non-linear currents within the output power transistors. So, to avoid all problems associated with class AB, we choose the less effective class A principle. Ideally, the power buffer is placed in close proximity to the loudspeaker to minimise cable/loudspeaker related error signals. The block schematic below illustrates our basic system concept.

Figure 1.

The connection between the two parts can be done with ordinary phono cables. The voltage amplifier has a higher gain than an ordinary preamplifier, since it at least should be the same as the voltage amplification in an ordinary power amplifier; and maybe some decibels more. As you can see from the block diagram above, if you exclude the input selector and volume control, you will have an ordinary power amplifier. However I would still recommend that the voltage amplifier is geographically remote from the buffer amplifier and has local optimised supplies.

Feedback, or no feedback?

Some experienced listeners report favorably on the sound quality of non-feedback amplifiers. This is surprising, because such amplifiers have more nonlinear distortion than amplifiers that use negative feedback. Indeed, the appropriate use of negative feedback improves almost all of the theoretical and measurable parameters of an amplifier. This has led to the suspicion that some subtle consequent of negative feedback may
be responsible for the difference in perception.

Of the various proposals attempting to explain how these perceptual differences might arise, most have suggested bad design errors in the application of feedback. For example, 'transient intermodulation distortion' (TID) occurs in amplifiers with inadequate slew rate. Transient intermodulation distortion occurs when the amplifier cannot slew fast enough to follow the transient. During such a transient, the amplifier is pinned to the slew trajectory and cannot follow variations in the input. Apparently this effect was known as early as Roddam in 1952, but it became widely known in audio circles with the work of Otala at the beginning of the seventies.

James Boyk (see Life above 20 kHz) has made a survey of wideband spectra from real musical sources. Although some of these show significant energy above 20 kHz they put limits on the rate of change of the sound pressure level in most ordinary music waveforms. For the CD media, the maximum frequency is about 20 kHz, and even for other digital sources, the maximum frequency is about the double. The Moving Coil pick up is the signal source where it has been reported some energy above 20 kHz . But real-world transients have rise times and signal levels that are easy to accommodate with modern circuits, to avoid TID.

Feedback amplifiers must be compensated to ensure stability, and the most common compensation scheme introduces a principal pole at low frequencies that lowers the loop gain as the frequency increases. This means that the feedback is less and less effective, the higher the frequency. The implications are numerous, some more serious than others. Firstly, the tonal balance of a musical instrument is given by the relation between the first harmonic and the second, third harmonic etc. The result is that the distortion is lowest for the first harmonic, and higher for the other harmonics; implying that the tonal characteristics of the instrument is changed; a negative consequence, indeed. But it is possible to design a feedback amplifier with an open-loop bandwidth up to 10 kHz (or more), as in my own designs, to overcome this problem.

Another implication is that the output impedance of a feedback amplifier may be very low at audio frequencies, but rises as the frequency increases. This may have as one consequence that these amplifiers are more sensitive to radiofrequency interference. If the impedance is high enough, strong radio-frequency fields can get in through the output and thus change the bias conditions of the amplifier. This is certainly possible, but a well-shielded amplifier with appropriate filters need not have this problem. However, when the output impedance raises with frequency, the loudspeaker sees a higher impedance than designed for.

Yet another idea is that the clipping behavior of feedback amplifiers is different from that of non-feedback amplifiers: clipping is sharper and recovery from clipping may be problematic. This also is true, but it does not explain the perceptual differences that may remain even in the case that the amplifiers are not driven to clipping: listeners report differences in the low-level details and the sound of the "room".

The possibility that the difference in sound quality is not an accident of the particular design, but is an inherent characteristic of negative feedback, has been proposed by many listeners. In 1957, Norman Crowhurst in 'Journal of the Audio Engineering Society', observed that since the intrinsic nonlinearity of an amplifier must produce harmonic and intermodulation products from the components of the program material, feedback will combine these products with the program to produce further distortion products. Since many of the products in each "generation" are higher or lower in frequency than the signals that produce them, the effect will be to create products extending over the full bandwidth of the amplifier. Although the total amount of this distortion is very small, much smaller than the lower-order distortion produced by the same amplifier without feedback, Crowhurst observed, "The logical result of this process would be a sort of program-modulated, high-frequency 'noise' component, giving the reproduction a 'roughness'."

Some has argued that this 'noise', since it is correlated with the program material and thereby is changing all the time, may interfere with the listeners perceptions of low-level details. Notice that this is equivalent to the problem with quantization noise in CD players and other digital sources. As far back as 1950, D. E. L. Shorter published "The influence of High-Order Products in Non-Linear Distortion," in Electrical Engineering. It has been confirmed that higher distortion products are more severe than lower, i.e. it is better with 0.5 % second harmonic distortion than 0.3 % third harmonic distortion.

However it is not possible (to my knowledge) to design a completely non-feedback amplifier. So here I use the term to denote an amplifier with no global feedback; in fact this amplifier has no feedback loops in neither the Voltage amplifier nor the Buffer amplifier. This means that the reduction in distortion is relying on local series feedback only.

The Buffer Amplifier


The current amplifier/buffer, view the circuit diagram below for the printed circuit board, is fully symmetric built up as an ordinary Darlington emitter follower. It is, when working in Class A, both inherent stable and very linear, at the same time having good thermal stability. The bias generator consists of the components 109-115, where it is used two transistors instead of the usual one to form a variable zener. The offset cancellation for this buffer is done by means of the potentiometer P114. Be aware that the offset adjustment should be done with shorted input.

Figure 2.

The quiescent current is set by the ratio of R112 to R115 in the negative leg (and R111 to R113 + P114 in the positive leg). The current from Q107-108 therefore also is a part of the quiescent current setting, since it detemine the base-emitter voltage of Q109 and 110. To change the quiescent current, increase R111 and 112 for increased current, and vice versa. The input impedance of this buffer is about 16 kohm for an 8 ohm load. The output impedance is nearly resistive througout the whole audio range, only barely inductive at the highest frequencies (a few degrees at 100 kHz).

Power Supply

The power supply for the Buffer amplifier can be made in many different ways. My proposal is a common transformer for the two channels when the Buffer amplifier is made as a stereo version. When a common transformer is used, it is advisable to use separate rectifiers, i.e. one for each channel, see the figure below.

Figure 3.

It is also advisable to use decoupling capacitors (100 nF) in parallel with the main filtering capacitors C1-4. The value of C1-4 should at least be 10000 µF, 22000 µF should in most cases be sufficient. The transformer should be a 2x18 V toroid for a 30 W per channel amplifier, and the rectifiers should be at least 25 A. This will give about +/- 25 V DC to the Buffer amplifier. The fuse F1 is somewhat dependent on the size of the filtering capacitors, but a good starting value could be 4 Amperes.

The Voltage Amplifier


The input stage of the voltage amplifier consists of two complementary JFETs in a symmetrical coupling. The offset adjustment is made up by the potentiometer P17. The resistors R1 + R2 set the input impedance. The resistor R1 will together with the output resistance from the source and the input capacitance of the amplifier set the upper cut-off frequency. If the source has e.g. 600 ohms output impedance, the cut-off frequency is about 3 MHz, giving an equivalent input capacitance of about 30 pF, but observe that the connecting cable will increase this capacitance.

Figure 4.

The field effect transistors J15 (N-channel) and J16 (P-channel) withstand a maximum of respectively 40 V and 20 V to operate properly, and their internal capacitances are high. Using the cascode coupling (Q13/Q14) as shown, solves these problems, and at the same time reduces the leakage current. This makes it possible to have a very high input impedance. The common base transistors Q13/Q14 are bipolar types. It is also used 'base degeneration' by means of the resistors R3-6 instead of using a constant voltage on the bases of the transistors.

Instead of using an ordinary current mirror for the current from Q13/Q14, it is used a very accurate four transistor current mirror in each leg (Q9/Q11/Q21/Q23 and Q10/Q12/Q22/Q24).

When the voltage on the input of the amplifier is transformed to a proportional current in the JFETs, this current is very exactly replicated in Q23 and Q24, and summed over R25 to give a voltage higher than the input voltage.The distortion at this point is very low compared to the relative high (even order) distortion from the field effect transistors. The ratio between this sum voltage and the input voltage is the voltage gain. This gain is set by the relation between R25 and R18 (plus ideally half of P17 and the equivalent resistance in the JFETs). With the values shown, the voltage gain is about 33 times (30.4 dB). About 0.45 V RMS input voltage is thus required for full output power (30 W). This should be sufficient for the most modern signal sources. If higher gain is wanted, you may reduce the value R18 and/or increase the value of R25. None of the good properties of the amplifier, like bandwidth, distortion and Slew Rate, are deteriorated by moderate change of R18/R25. If you want to use the voltage amplifier only as a preamplifier and/or a headphone amplifier, a natural choice would be to reduce the value of R25 to the half, as an typical example.

The components 27-41 form a Darlington emitter follower, where Q32/33 set the current of Q36/37 and finally the current of the output transistors Q38/39. I have used medium power transistors for these, making it possible to use this voltage amplifier also as a headphone amplifier. If you do not want this, you may replace these transistors with small signal types and increase the value of R40/41 accordingly. The use of this Darlington output stage ensure low impedance drive to the Buffer amplifier. This makes it possible to use long cables between the voltage amplifier and the the Buffer amplifier. The purpose of R26 is to reduce the voltage gain when the voltage amplifier is used as an ordinary preamplifier or as a headphone amplifier. The relay K99 makes it possible to lower the gain by grounding X9 (RL).

The amplifier has a bandwidth of about 1 MHz. The harmonic distortion is very low to be a non-feedback design, and it is nearly independent of the frequency in the audio band. The output impedance is resistive througout the whole audio range. Since it is no compensation capacitors in this amplifier, the Slew Rate limitation has been difficult to measure, but it is very high (at least 200 V/µs with the input stage driven into saturation).

DC Servo

The output from the voltage amplifier should have no DC offset. Therefore I have added a DC servo, se below. However, this servo has a current output, not voltage output as usual with servos designed by use of operational amplifiers. When the amplifier output (OUT) goes positive, this increases the current in Q92. This, in turn, means that the voltage across R18 (in the voltage amplifier) is forced positive; decreasing the current in J15 and increasing the current in J16. In this way DC balance is restored.

Figure 5.

J85 and J86 form a differential pair, while J89 forms a current source. R90 determines the current in the differential pair. Q83 and Q84 is a current mirror, while Q92 is an ordinary common emitter gain stage. R79/C80 and R94/C95 determine the lower cut-off frequency. Via the resistor R96, the DC current in the Voltage Amplifier is adjusted. C97 is used to ensure amplifier stability and is soldered directly on base-collector of Q92, on the solder side of the printed circuit board.

Power Supply

The on-board power supply schematic is shown below. The center-tapped transformer (not shown) is connected to the AC inputs and GND input. Following the rectifier D97, the filtering capacitors C42 and 43 smoothe the rectified voltage. J46 and 47 form current sources for the low noise references D50 and 51. The voltage from the references are filtered by means of the components 54-59. The components 60-73 form a voltage regulator for positive and negative power supply voltages for the voltage amplifier and DC servo. These voltages is set by the reference and the voltage dividers R70/72 and R71/73. To increase the output voltage to more than about 28 V, the voltage with the values shown, you may increase R70 and 71. You may use the formula V+REG=6.8(1+R70/R72) for the positive regulated output voltage (and similar for the negative output voltage). Together with the additional filter capacitors C74 and 75, the regulators give a low noise output voltage with very little ripple.

Figure 6.

It is advisable to place the transformer in the Buffer amplifier enclosure; in that way the magnetic fields are not influencing the voltage gain stages. In addition it is possible to also place the rectifier and filtering capacitors in the Buffer amplifier enclosure. Then one can use the solution in Figure 3, when a common transformer for the two channels is used. In that case, the rectifier D97 and capacitor C98 in the schematic above may be omitted, and the positive and negative voltage is connected to the V+ and V- terminals instead. I would, however, recommend to still use the capacitors C42 and C43 on-board. This gives a further reduction in ripple and noise. The power supply voltage for the Voltage amplifier should be a few volts higher than the power supply voltage for the Buffer amplifier, at least 3 V higher.

Mounting description

The parts list applies for both the circuit boards and the power supply.

The Buffer Amplifier

The printed circuit board (PCB) measures about 137x39 mm, see the component placement for both the Buffer and the Voltage Amplifier below.

The layout is shown below.

With the exception of the power resistors, 1/2 W metal film oxide resistors with 1 % tolerance may be used. The power resistors in the list are non-inductive, and alternatively they may be realized by connecting power metal film oxide resistors (e.g. 2 W) in parallel. I have used ordinary 3 mm red LEDs for D103/104. A replacement for the couple 2SC1775/2SA872 used for Q107/Q108 is e.g. 2SC1815/2SA1015 or 2SC2240/2SA970. 2SC3421/2SA1358 is used for the bias generator and the driver. These may be replaced by 2SC4793/2SA1837. For all replacements, be sure that the pinning is correct when mounting the transistors on the circuit board.

The output transistors used for Q121/Q122, are the well-known couple 2SC2922/2SA1216 from Sanken. If you are looking for a replacement for these, be sure they are complementary and have the necessary power rating. The couple comes in a rare plastic housing, something that makes it necessary to mount them directly to the heat sink. This is also done for the drivers (Q118/Q119) and the bias transistors (Q109/Q110). It is also a good idea to mount the PCBs directly on the heatsinks.

The outputs from the transformer are carried to the rectifiers as shown in the power supply schematics in Figure 3. Optionally, you may use fuses here in case of the rectifiers break down. You may use the principle of star ground here; that is, use a common connection between the filter capacitors (C1-4) to chassis, where you also connect the mid point from the transformer. The phono socket ground terminal is connected to chassis (near the input) and the (ground) shield of the phono cable is not connected to the PCB ground. The inner conductor of the phono cable is connected to the PCB point marked with 'IN'. From the loudspeaker output the two conductors are twisted and fastened to the PCB in the two points marked with 'OUT' and 'GND'. The last one is connected to the minus conductor. All connections should be as short as possible. If some sort of instability or noise should occur, the probability is high that the reason is bad wiring (e.g. earth loops).

The Voltage Amplifier

The printed circuit board (PCB) measures about 137x75 mm (The layout and component placement are shown above). All the resistors can be 1/2 W metal film oxide type with 1 % tolerance. It is always a good idea to start with the passive components. However, do not mount the resistor R96 in the DC Servo (Figure 5). The JFETs 2SK170BL/2SJ74BL should be used for J15/J16. The value of P17 is chosen to give a current of 5-7 mA from the JFETs. The output transistors 2SC3421/2SA1358 is used for Q38/Q39. These may be replaced by 2SC4793/2SA1837. The resistors R40/R41 set the quiescent current of these transistors to about 20 mA with the values shown. Be aware that if you choose to increase the quiescent current, it may be necessary to mount these transistors on small heatsinks. For the other bipolar transistors in the Voltage amplifier (Figure 4), I have used the couple 2SD756 (NPN) and 2SB716 (PNP). These transistors are 0.9 W types, but with similar properties as 2SC1775/2SA872 shown in the schematics.

In the DC Servo (Figure 5), I have used the bipolar transistor types shown in the schematics. Here it is possible to use all the couples 2SD756/2SB716, 2SC1775/2SA872 and 2SC2240/2SA970. In the on-board Power Supply (Figure 6), I have used the couple 2SC1815/2SA1015. These are quite linear for higher collector currents, but can be replaced by the other couples mentioned earlier. For all replacements, be sure that the pinning is correct when mounting the transistors on the circuit board. In the DC Servo (Figure 5), remember to mount the compensation capacitor on the solder side between base and collector of Q92.

The other components should be quite straightforward, but be sure to use the LM329 as a reference, since this generates much less noise than an ordinary zener diode. The JFETs in the DC Servo (Figure 5) and in the on-board Power Supply (Figure 6) are not critical, but I have used 2SK170BL. If you place the transformer, rectifier and filtering capacitors in the Buffer Amplifier enclosure, following the schematics in Figure 3, you should be aware of possible ground loops. One simple way to reduce this risk, is in the Voltage Amplifier to use a 10 ohms resistor from the power ground inlet to chassis (See Figure 6). This will establish a voltage difference between ground in the Buffer Amplifier and the Voltage Amplifier.

If you are using the concept in Figure 1, you may choose a value on the volume potentiometer P1 as you like. And if you prefer to have a remote controlled volume, you may choose an Alps type of potentiometer. The number of inputs is of course your own choice. If you really want a Tape output, you may consider the loading of this (or withdraw the connection when it is not in used). The relay K99 is activated when RL (X9) is grounded. If the unregulated voltage is much higher than 24 V DC, you will need a series resistance to ground. The easiest way to activate the relay (K99), is to use a switch. If you do not need a headphone output, you will not need the relay (and resistor R26).

Start-up and adjustment

The Buffer Amplifier

It is recommended to use a variable transformer or variable DC voltage generator first time the amplifier is started up. When the power supply voltage is increased, adjust the output-offset voltage by means of the potentiometer P114 to be close to 0 V DC with the input shorted. Using the values in the parts list, the amplifier will perform 30 W RMS into 8 ohms. This demands a minimum quiescent current of 1.4 A for class A. Heatsinks with a minimum capacity of 0.35K/W is recommended. Control that the quiescent current is about 1.8 Amperes. I regard this to be about optimum for a 30 W amplifier. The resistors R111 and R112 are used to change the quiescent current, lower values give lower current. If the quiescent current is increased, do remember the cooling demand.

If possible, look at the output with an oscilloscope, there should not be anything but very little noise here if everything is OK. When the temperature is increasing, it may be necessary to re-adjust the offset. The offset voltage normally varies, but should not exceed 100 mV. The quiescent current should stabilize when the heat sinks reach their working temperature. If this continues to raise after an hour, it is something wrong. If you want it, you may test this amplifier with a signal generator. For 30 W RMS output power, the voltage needed on the input is about 21.8 V peak, and the load is 8 ohms.

The output power of this amplifier may be increased to 50 W RMS in class A. This can be done by using a toroid transformer of 2x22 V instead of 2x18V. Then the quiescent current should be increased to 2 Amperes or more. The cooling requirement, however, is large and should not be underrated. Please notice that a doubling of the output power not necessarily give a subjective feeling of a more powerful amplifier, but a class A amplifier is generally perceived more powerful than a class B (or A/B) amplifier.

The Voltage Amplifier

Start with connecting the supply voltage to the PCB, and control the regulated supply voltage. This should be about 28 V for the component values shown. If you have the possibility, you may check these outputs to see if there are any ripple and noise. The next point is to check the output of the DC Servo, at the junction of Q92 and R93 without R96 mounted. Apply ground to the junction of R79 and C80 (at the input of the DC Servo), and check the output of the DC Servo. If this voltage is many mV from exactly 0 V, you may change the value of R81. Normally it should be sufficient to change this value only slightly. Then be sure to remove the shorting (to ground).

After this is done, adjust the offset potentiometer P17 such that the output of the Voltage Amplifier is close to 0 V DC. Now, mount the the resistor R96. The output of the Voltage Amplifier now should be only a few mV away from 0 V DC. You should check the voltage across R40 (and R41), it should be about 0.45 V. If you want to use higher output current here, be sure that the transistors (Q38/Q39) do not run too hot.

The complete amplifier

Now it is time to test the complete amplifier, i.e. the Voltage Amplifier followed by the Buffer Amplifier, with a signal generator. For 30 W RMS output power (about 21.8 V peak output voltage across 8 ohms load), the voltage needed on the input is about 0.45 V RMS. If you have the possibility, you may test the amplifier for distortion, bandwidth, Slew Rate etc.

Some measuring results

Input impedance:
Output power:
Output impedance:
Bandwidth (small signal)
Slew Rate:
THD at 10 V RMS:
Rise/Fall time:

0.45 V
220 kohm||30 pF
30 W RMS
< 0.2 ohm, 2 degrees (1Hz-100kHz)
DC-1.2 MHz
< ± 200 V/µs
0.02 % (200 Hz, 1 kHz and 5 kHz)
< 1 µs

Please notice:
This project description is for non-commercial use, only. Using this document on a site and charging a fee for download is vialation of non-commercial use and prone to demand for payment. So, for commercial use, contact me for agreement of terms. This page, however, can be downloaded for own use, and linked to, not violating term of non-commercial use.


Knut Harald Nygaard