Correlation Radiometer
How Cross-Multiplication Cancels Uncorrelated Noise
A square-law radiometer measures power by squaring a single amplified signal, so its output carries the full system noise temperature as a large, fluctuating pedestal that must be stripped away by Dicke switching or load comparison. A correlation radiometer takes a different path: it carries the signal through two parallel amplifier chains and multiplies their outputs in a correlator. If the two inputs share a common component, such as the same antenna voltage split into both arms or a single injected reference, that component appears in the product as a real, slowly varying term. The independent thermal noise added by each amplifier, by contrast, has no correlation between channels, so its contribution to the time-averaged product converges toward zero.
This hardware cancellation is what makes the architecture attractive for polarimetry and interferometry. In a radio interferometer the two chains are fed by two physically separate antennas, and the correlated output is the complex visibility that encodes source structure. In a polarimeter the two chains carry orthogonal polarizations, and the cross-correlation directly yields the Stokes Q and U parameters. In both cases the correlator output is naturally referenced near zero, so there is no large DC term and no need to subtract a hot or cold reference load on every cycle.
The penalty is a modest sensitivity factor. Because two noisy channels contribute fluctuations while only the correlated power is retained, the radiometer equation constant rises to roughly the square root of 2 compared with an ideal total-power receiver. Designers accept that small loss in exchange for immunity to the slow gain drift and 1/f noise that dominate long integrations, since uncorrelated gain ripple between the two chains also averages out of the product.
Governing Equations
<v1(t) × v2(t)> ∝ √(G1G2) × κ × kBTcorrB
Radiometer sensitivity:
ΔT ≈ k × Tsys / √(B × τ), k ≈ √2
Required integration time:
τ ≈ (k × Tsys / ΔT)2 / B
Where G1, G2 = power gains of the two chains (correlated output scales as their geometric mean), κ = correlation coefficient (0 to 1), kB = Boltzmann constant, Tcorr = correlated antenna temperature, B = predetection bandwidth, τ = integration time, Tsys = system noise temperature. Example: Tsys = 50 K, B = 2 GHz, target ΔT = 0.1 K → τ ≈ 250 μs.
Radiometer Architecture Comparison
| Architecture | Detector | Sensitivity k | Gain-drift immunity | Output pedestal | Typical use |
|---|---|---|---|---|---|
| Total-power | Square-law | 1.0 | Poor | Full Tsys | Calibrated power monitor |
| Dicke-switched | Square-law + switch | 2.0 | Good | Near zero | Ground microwave sounders |
| Correlation | Multiplier | √2 (1.41) | Excellent | Near zero | Interferometers, polarimeters |
| Complex correlation | I/Q multiplier | √2 (1.41) | Excellent | Near zero | Visibility / Stokes mapping |
| Noise-injection | Square-law + ref | 1.5 to 2.0 | Very good | Stabilized | Spaceborne calibration |
Frequently Asked Questions
How does a correlation radiometer cancel receiver noise that a total-power radiometer cannot?
The two amplifier chains carry statistically independent internal noise, so when their outputs are multiplied and time-averaged the cross term between each amplifier's own noise converges to zero. Only power common to both inputs, a split antenna signal or shared reference, survives. A total-power receiver squares one chain, leaving the full Tsys of 30 to 80 K sitting on the output as a pedestal that switching must remove. The correlator removes that pedestal in hardware, which is why it dominates interferometers and polarimeters.
What integration time does a correlation radiometer need to reach 0.1 K sensitivity?
From the radiometer equation, ΔT ≈ k Tsys / √(Bτ) with k near √2. For Tsys = 50 K, B = 2 GHz, and a target ΔT of 0.1 K, the required time is τ = (k Tsys / ΔT)2 / B, roughly 250 μs of correlated averaging. Faint astronomical sources are integrated for seconds to hours to push into the millikelvin range.
Why is a cryogenically cooled LNA used ahead of the correlator?
The correlator only suppresses the uncorrelated part of each amplifier's noise; it does not lower the system noise temperature that sets sensitivity. So the first stage in each path is a cryogenic LNA cooled to 15 to 20 K, giving a noise temperature of 4 to 10 K at X and Ku bands. Cooling also stabilizes amplifier gain, reducing the slow drift that would otherwise leak into the correlated output.