school Sep 2025

Building an op-amp from transistors

A discrete operational amplifier, built up from a handful of BC547 NPNs and BC557 PNPs on a breadboard, powered at ±5 V, then measured open-loop, closed-loop, and under two different loads. The point was less to use the op-amp than to see it from the inside.

Hand-drawn schematic split into two halves: 'Differensiell forsterker' on the left and 'Buffer' on the right. The differential half shows T1/T2 forming a current mirror at the top, T3/T4 the PNP differential input pair, with a tail-current source built from T5 biased by R1 (47 kΩ) and R2 between V+ and V−. The buffer half shows T6 as an emitter follower into a 1 kΩ source resistor R_S, with output v_o. — The full schematic: differential pair, current mirror, tail source, emitter follower.

The architecture

Differential input pair (T₃, T₄): inputs land on the bases of a PNP pair; the tail-current transistor (T₅) sets the bias.
Current mirror (T₁, T₂): turns the differential current into a single-ended signal.
Emitter-follower output (T₆): low output impedance so the op-amp can actually drive a load.
Bias network: R₁/R₂ voltage divider sets the tail transistor's reference. R_S = 1 kΩ at the emitter follower limits the output current.

What I measured

1 kHz sine, ±5 V supply, two loads (100 kΩ and 100 Ω), open-loop and with a ×10 inverting feedback network:

Config	R_L	V_in	A	THD_out
Open loop	100 kΩ	1 mV	~400	4.8%
Open loop	100 Ω	3 mV	~100	1.7%
×10 feedback	100 Ω	50 mV	9.2	0.27%
×10 feedback	100 kΩ	50 mV	9.7	0.22%

Open-loop gain crashes by a factor of four when the load drops from 100 kΩ to 100 Ω. The discrete output stage cannot keep up. Closed-loop, with the gain nailed to a target of 10 by the feedback network, both load values land within a few percent of the target and the THD drops by an order of magnitude. Feedback does exactly what the textbook says it does, on real hardware.

The discrete op-amp built on a stripboard. Six small-signal transistors crammed into a corner with a tangle of red, yellow, orange and black wires running to the breadboard underneath. — The build. Six transistors, one stripboard, more wires than expected.

Two overlaid FFTs of the output: orange trace (open loop) sits noticeably higher than blue trace (×10 feedback). THD labels in the corner: 0.223% open loop vs 0.6633%; actually the closed-loop version is the blue, much cleaner, lower-amplitude noise floor. — FFT, open-loop (orange) vs ×10 feedback (blue). Same fundamental, but the harmonic floor drops about 20 dB once feedback is closed.

What I took away

That seeing closed-loop feedback dominate component variability with your own eyes is more convincing than any number of textbook diagrams. The 9.2 → 9.7 spread between loads, and the 9.2 → 10.7 swing I got when swapping one feedback resistor for a nominally identical part, made the case for negative feedback better than any lecture.

Writeup

Download: designnotat_6.pdf (Operasjonsforsterker)

Back to timeline