The amp (see above and pdf-files) has very low noise (if you equipt the amp with low noise transistors) and an extremely flat frequency response, up in the megahertz-region. The input impedance is determined by the input bias servo. With the chosen resistor values the input impedance is 200 kohms.
In order to get an extremely linear and extremely low distortion design I have taken a couple of measures:
- Symmetrical design
- No capacitors in the signal paths
- Cascodes in the input stage and in the high gain stage
- Buffered cascodes with emitter followers
- Possibility to use ultra low noise transistors
- High speed output stage of four different kinds, small MOSFET's, larger ones, and bipolar transistors of virtually any kind as long as they are of TO220 or TO126 type.
- Input bias servo
- Totally DC-coupled together with DC-servos and input bias current servo
- Many decoupling capacitors and EMI-filters
- Built-in power supply including transformers
- Professional quality printed circuit board with groundplane
The whole amp is completely symmetrical (not entirely because of the fact that NPN and PNP never will be their exact opposites) with help from PNP- and NPN-transistors. The benefit of a symmetrical design is lower distortion and I like symmetrical designs, just by the looks.
No capacitors in the signal paths
Two DC-servos eliminates the input offsets, which always are present. No capacitors are the signal path. Therefore the output offset voltage is about 10 times higher than in a normal AC-coupled amp. The residual offset comes from the DC-servo and the opamp. The output offset voltage is the DC-voltage at the output when you have no signal applied to the input. It comes from deficiencies in the input stage times the total DC-gain. In my case it is some millivolts times the closed-loop gain. The offset comes from differences in the input transistors characteristics, temperature, DC-gain etc. It's impossible to make two identical transistors. There is always a difference between two transistors. You can minimize the defect by having two or more transistors on the same chip. Examples of very good matched monolithic transistor pairs are SSM-2210 (not available anymore) and same type in metal can MAT02 (NPN) and the PNP-types SSM-2220/MAT03. The difference is only DIL08 plastic package and TO-78, metal can....and the price I must add!.....1,5 vs 10 dollars! I have tested with good results the ultra tiny BC847/BC857BS but these transistors are not suitable for newbeginners because thet are so tiny. See the picture!
Is there anyone who knows if there are alternatives to MAT02/03?
Input bias servo
To reduce the output offset voltage further the amp has also a circuit which injects DC-current to the input. The input currents are not negligible if ultra low noise transistors are used. If you are going to use normal BCxxx transistors there is no need really for this servo but I doesn't hurt to have them.
The cascode stage is a very useful combination of two transistors. The benefits are two, lower distortion and dramatically extended frequency response. T15 and T17 is one of the cascodes. The gain is 20000(!) and the bandwidth 20 kHz. Maybe it's not that interesting to know why the stage gets faster, but it's true. But why does the distortion decrease?
The bipolar transistor is a current driven device. You are only interested in input (base) current and output (collector) current. The picture above shows that the current gain is dependent of the applied voltage on the collector. If i.e. Hfe is 300 at 1 mA and 25V, it gets 350 at 35 volts. The current gain isn't a constant parameter. It changes also if you change the base current. 10 uA base current generates 1 mA, 20 uA, 2,2 mA, 30 3,6 mA. The fact that the current gain isn't constant generates harmonic distortion. But the greatest "defect" is the sensitivity for varying collector voltages.
If you take out the output signal at the collector with a help from a collector resistor, you'll get a signal that moves along the load line, see above. The graphical way of showing it is that the lines for Ib1, Ib2 etc. gets more separated at higher collector voltages, at the right. This is a sign that the current gain is higher than at the left (low collector voltages). Parallel, but maybe inclining, lines is a sign of constant current gain.
If you try to keep the collector voltage constant and in some way try to use the collector current you get interesting results. The "magical" load is the input of a base grounded stage. The base is at a constant potential with respect to signal ground. The input impedance of a such stage is very low, 25 ohms at 1 mA, 2,5 ohms at 10 mA. The base grounded stage isolates the first transistor from collector voltages variations.
The results of this are:
- Elimination of the Miller effect and therefore increases (extends) the frequency response dramatically
- Reduction of distortion
- Creates a super transistor with high gain, high voltage and high speed and also high power in some circumstances.
Advantage to use cascodes in an input stage:
- Faster input stage and therefore easier to feedback and to get a stable combination. The input stage should be as fast as possible.
- Possibility top use low voltage and low noise transistors and high supply voltage, especially important in power amplifiers.
Advantage to use cascodes in a high gain stage:
- Possibility top use low voltage and fast transistors together with high voltage ones and achieve high gain at high voltage swings.
- Low distortion
- Faster stage (sometimes an advantage)
Explanations of the specific design
Please download all the schematic pages and print them out. It's much easier to understand what I'm talking about if you look at the circuits at the same time. I will only talk about the left channel and the upper half. The design is symmetrical and the two channels are identical.
The input filter
The input filter is for "just in case" but still it's very wise to have it. If you choose to use the input bias servo the input impedance will be 100 kohms (R3) and if you choose not it's several megaohms (you can ignore it).
R1 is a pulldown resistor if not coupling caps (C2, C3) or input bias servo are used.
R2 together with C1 is an EMI-filter. In order to achieve the lowest noise possible (not very important in this application) R2 should be 0 ohm if you don't have any RF interference. I recommend that you always has this filter.
The differential (input) stage
Current source for the stage is DZ2 sets a 6.8 V voltage, fed through a LP-filter (R11, C9, C11) with reduces noise from zener and power supply and makes the amp start up nice when the power is switched on. The voltage drop across R11 is approx. 0.7 V. This makes a voltage across R13 of 5.56V. The current is therefore 3,7 mA divided into two parts feeding the differential transistors (T5, T9) which can be separate components or monolithic and matched pairs.
T7 and T11 forms a cascode together with the lower transistor pairs, T5 and T9. This type of stage extends the frequency response by a factor of at least 10. The distortion is also reduced. T3 creates a constant current and a constant voltage across R16 related to ground. The voltage is approx. 5.5 V. T7 and T11 are biased via R18 and smoothed by C11 and C14.
In order to get higher speed of the input stage, the emitters of T5 and T9 have resistors (degenerated emitters). This reduces also the gain and like a bonus also the distortion.
The gain is (due to the emitter resistors) R22/(R20*2) = 6, 15.5 dB.
The noise performance is the best at 100-300 µA with BC550/560 transistors. 1 mA or more is better for speed (easier to avoid oscillations). If you use MAT02/03 the optimum current is approx. 1-5 mA. If you choose to use MAT02/03 I recommend 1 mA through each transistor. Don't forget to recalculate the associated resistors.
The emitter follower after input stage and cascode
The emitter follower T13 decreases the effect of the load from the high gain stage T15 and T17. The result is mainly extended frequency response of the first stage. The second result is lower distortion. The main advantage of the emitter follower is increased phase margin and therefore better stability when the whole amp is feedbacked. The odds not to have oscillations increases.
The frequency response of the emitter follower is very high (>> 10 MHz) and also a little bit tricky to measure it.
The high gain stage
The important high gain stage is formed by T15 and T17. A minor local feedback is created by R30. The total gain is quite extreme, approx. 20000 times! The gain is lowered further with R32. The result is gain determined by R30/R32 = 1000 (60 dB). C20 creates a dominant pole in order to get a stable amplifier when feedbacked. The gain is very easy to adjust, just change the relationship between the mentioned resistance's. It's possible the lower the capacitance's in order to get a faster stage and lower the distortion. An easy way to test how small the capacitance can be, apply square wave and trim for a perfect step response. The value of C20 is a compromise in order to have speed but no oscillations. Chosen value is safe in order to avoid unstability, no gambling.
The fast output stage
I have chosen four different kinds of power transistors.
- DIL-04 small cute MOSFET's
- GDS, Ordinary pinned power MOSFET's
- GSD, "Hitachi" pinned power MOSFET's like 2SJ79, 2SK216
- ECB, TO126 types like BD139/BD140
The small IRFD120/9120 are my substitutes for the orginal Supertex MOSFET's but I figured that much that it would be smart to design a 100 % universal output stage. These IRFD120/9120 are kind of odd with it's package DIL-04. You will get good performance of every kind of output stage. It's more like a matter of taste.
The bias circuit
The bias circit is made of a common Vbe multiplier and the transistor (BC550 like) is either glued to one of the IRFD120 fet's or mounted on heatsink but then you may use BD139 as sensing element.. If you use MOSFET's is't an advantage to use a MOSFET as sensing element because temperature coefficient will be more alike. If you use a bipolar transistor with MOSFET's you will get an amp which has high start current. No harm in that but it's maybe nicer to have constant current.
Unfortunately the class A temperature regulation circuit needs more development. Forget it for now! The main problem with the regulator is current limiting. You must sense the current somehow.
It's virtual impossible to get good DC-performance without some kind of servo (based on good opamps). The servo has no problem to be connected conventionally to the junction of the inverting input (base of T9, T10) but I prefer the chosen technique.
I have chosen to inject current inside the amp instead. Through R60, R61 a small current is either injected or drained. C37 makes sure that no audio signal goes into the servo and also that the servo not introduces any noise. The opamp IC2 should be a decent type with good DC-performance, slow and with low noise. The speed of the opamp should be not more than 1 V/µs. The opamp works as a non-inverting integrator. You should always have symmetrical component values in order to get a true integrator and also to get low offset caused by input offset currents of the opamp.
Input bias servo
Because the input stage has noticeable input current, an offset voltage is always present. If ultra low noise transistors are used, this servo is very important because of the high base current from the negative half. The PNP types of this transistor type has rather low current gain and the NPN-type rather high. This causes an unbalance in base currents. IC1 works a normal inverting integrator which strives to keep the output of the opamp at zero volts. The servo speed is very dependent of the signal source impedance. Very low source impedance creates very long time constants. If the R2 of the input filter is decreased the servo gets even slower.
The total open-loop gain is 6000, 75.5 dB and the closed-loop, 11,(= 20.8 dB). The closed-loop gain is set by (R49/R51) + 1, a little bit high maybe but it can be decreased without any problems. 1-10 is normal gain values. My Sennheiser HD545 needs 2-4 times in gain for 1 V signal sources in order to play real loud.
SMD types, alternative transistors
Long time ago (in 1986) when I first designed this amp it was very easy to get good hole mounted types of transistors but now it can be quite hard to get good transistors with high gain. It's rather easy though to get good surface mounted types. Therefore I have made room for SMD types in every transistor position. I have also included the possibility to use cheap transistor pairs (very small devices) and for matched ultra low noise matched pairs. SMD transistors aren't so hard to solder by hand but BC847BS/BC857BS are not suitable for newbeginners, only if you are VERY handy with the soldering iron.
Rescently when I was soldering the prototype I discovered that the video transistors BF471/BF472 are out of production since a couple of years! I can't find any TO126 type with similar characteristics (extremely low Cbc). You can use ordinary TO92 types but I have made any room for these. You can mount them in the TO126 holes but it doesn't look so nice. You could also you SMD transistors. They will fit.
... more to be said......Note that there is an option for using a low profile transformer or for a "normal" one.
I have nothing against "audiophilic" components but this design requires normal standard good performing industrial parts as a good start and will work correctly if the recommend parts are used. Feel free to change the parts you like but you must have control over the important electrical parameters of each part. I think that some of the "audiophilic" parts are just plain froud. They have absolutely no documented "audiophilic" features what so ever. Same parts are only known but rumours over the internet and magazines. In many cases no double blind tests have taken place.