Curiously Low Noise Amplifier

Curiously Low Noise Amplifier

The Curiously Low Noise Amplifier takes advantage of the wonderful noise characteristics of the 2SK170 JFET that boasts a noise voltage below 1 nV/root-Hz and virtually no noise current. The noise voltage of the amplifier is only 1.4 nV/root-Hz at 1 kHz, increasing to only 2.7 nV/root-Hz at 10 Hz. The noise current is difficult to measure, so this simple utility amplifier can see the noise from a 50 ohm resistor and a 100k resistor, too. (The 1.4 nV input-referred noise will increase to about 1.7 nV with a 50 ohm resistor, instead of a short, and a 100k resistor will give an input-referred noise near 40 nV, with very little contribution from the amplifier.)

This amplifier is a "utility" amplifier with a gain of 100, that would typically be used in a lab setting to boost tiny signals for measurement or further processing. It isn't intended to drive a speaker or headphones directly. (It could drive the LM386 quite nicely.) The circuit is a simple discrete transistor feedback circuit with two gain stages and a unique class-A output buffer:

• The 2sk117 is from the "BL" Idss current range and is selected for an Idss near 7 mA. The drain resistor is adjusted to achieve about 4 volts on the drain and the value depends on the Idss of the JFET.
• Most of the resistors aren't critical, but precision values are shown because the resistors should be metal film types for best noise performance. Approximate DC voltages are shown for helping with resistor selection. Deviating from the shown voltages will reduce the available output voltage swing, but the amplifier might work fine for smaller signals. Unloaded swing should be about 6 volts, p-p with about 60 mV p-p input, before distortion is observed.
• The MPSA18 acts as a noise filter. High gain is desirable here to keep the value of the base filter capacitor reasonable, but a 2N4401 could be substituted by reducing the 10k and 120k by a factor of 5. The filter will still be rolling off the noise voltage from the 15 volts supply above about 0.2 Hz. But some power supplies can be really noisy!
• The 0.1 uF capacitors serve as bypass capacitors but mainly as terminals for holding the components. These are the white rectangles seen in the photo.
• The feedback resistor is selected for a gain of exactly 100 and the value is well above the expected 1k, due to the limited open-loop gain of the simple circuit.
• A small resistor is included in series with the output for stability and that resistor can reduce the gain a bit when driving a lower resistance load. The designer may choose to set the gain for that particular load, say 75 ohms, or for a high impedance load. The circuit can drive a lower resistance than 100 ohms,  but the swing will be somewhat limited. It may be possible to leave out the 33 ohm resistor without stability issues. (Usually, such a utility amplifier is driving a much higher resistance load, typically 600 ohms or above.) Note: To give you an idea of how you can play with the output resistance, I just changed my unit's series output resistor to 55 ohms and adjusted the gain for 55 dB, when driving 75 ohm loads. Unloaded, the gain is exactly 5 dB higher at 60 dB. This way, I have even number gains whether driving a 75 ohm instrument or a high-Z device. The output buffer has no trouble driving the total 125 ohm load, with a swing limit of about 3.5 volts, p-p.
• The output stage is an unusual self-biasing arrangement where the PNP holds the gate-source voltage near 0.6 volts, running the JFET somewhat below its Idss. The 2N5486 was chosen to not waste too much current, but a higher Idss JFET will give more drive capability, if desired.

• Input Impedance: 47 megohm (set by bias resistor), shunted by 20 pF
• Output Impedance: 36 ohms, set by series resistor plus about 3 ohms from the circuit. My 55 ohm resistor mentioned above gives an output Z of about 58 ohms and exactly 5 dB of gain loss from no load to 75 ohms.
• Output voltage swing: 6 volts p-p into a high impedance load.
• Gain: 100 (40 dB) set by feedback resistor. Lower gain could be selected for wider bandwidth.
• Frequency Response: flat from below 1 Hz to above 2 MHz.
• Input Noise: 1.4 nV, rising to 2.7 nV at 10 Hz. Noise current has eluded measurement so far, but it's really low. With a 97.3 k resistor (100k in parallel with 3.6 meg) connected across the input, the noise voltage measures within a tiny fraction of a dB of 40 nV, so little to no noise current is seen. In fact, this amp and a selected resistor make an inherently accurate noise source. Connect a 152k across the input (in a shielded box), and you have a precise 5 uV/root-Hz noise source throughout the audio spectrum (50 nV times 100). A quick measurement at 40 Hz gives 770 nV/root-Hz with nothing connected; the 47 megohm is expected to contribute 867 nV. That's pretty close and still little noise current from the FET.

For even better performance, the bipolar stages could be replaced with a low noise op-amp. The input noise would drop a little, perhaps to 1 nV, as would the input capacitance, perhaps below 10 pf. Compensating the op-amp might be a bit of a challenge.