A radio wave reaches the antenna, not our ears

A peak appears on a radio telescope’s display. It could be a line from cosmic gas, a terrestrial transmitter, the instrument’s own electronics or processing. Explaining that peak requires knowing the entire path from incoming signal to plotted number, not just where the dish points.

A radio telescope is a receiving system whose antenna converts an electromagnetic field into an electrical signal, and whose receiver and processing measure its properties. A dish can help collect the wave, but it is not the only antenna shape. Radio means a part of the electromagnetic spectrum, not sound travelling through space. If a recording is rendered as hiss through a loudspeaker, the speaker creates those mechanical vibrations. NASA: radio waves and their conversion to sound.

Passive astronomical observing receives radiation from a source; it need not illuminate a star with its own transmitter first. Radar, using transmitted and reflected signals, is a different task, although some antennas have served both purposes. Here we discuss reception. The author has not built a radio telescope or measured signals from the sky for this article.

Investigating interference became a way to investigate the sky

Karl Jansky at Bell Labs investigated interference affecting transatlantic radiotelephone service. Part of the hiss neither remained in a fixed terrestrial direction nor followed the Sun. Its repetition relative to the stars became important evidence. A WJZ broadcast transcript dated May 15, 1933 describes the maximum arriving roughly four minutes earlier each day, following the stars’ daily motion. Something had not merely hissed: a testable relationship with time and direction had emerged. Jansky archive, broadcast transcript, pages 1–3.

Grote Reber followed up with a different design, building a parabolic antenna in 1937. In his 1940 paper Cosmic Static, he described holding a selected declination while Earth’s rotation carried different sky regions through the antenna beam. These records were not ordinary optical photographs. Reber himself noted that the instrument’s zero level drifted as it warmed up: receiver behaviour was part of the measurement from the beginning. NRAO timeline, Reber’s paper, pages 621–624.

Having a dish did not make the instrument equally sensitive to everything in the radio sky. Reber reported unsuccessful attempts at 3300 and 900 MHz; his successful data were obtained at 162 MHz. Frequency, receiver, source and sensitivity limit mattered. Construction in 1937, observations in 1939 and publication in 1940 are distinct events.

Dipoles, dishes and stationary arrays select directions differently

A changing electric field acts on charges in conductors. An alternating voltage appears at the antenna terminals and can be delivered to a matched receiver input. This does not mean that an individual electron travelled from a galaxy into the cable. The electromagnetic wave brings energy; charge motion within the material produces the electrical response.

A dipole has two conducting arms with a feed or receiving connection between them. Its directional pattern matters: sensitivity depends on arrival direction and electric-field orientation. It should not be described as equally sensitive in every direction. A dish collects the wave over a larger aperture and directs it toward the feed, which couples the field into the receiving chain. This may be a horn and waveguide or another arrangement appropriate to the band. NRAO: antenna and feed.

An antenna does not have to be a dish

Dipole

Dipole

Terminal voltage; sensitivity depends on direction and polarization.

Dish and feed

Dish and feed

The reflector couples a large aperture’s field to a smaller receiving assembly.

Stationary array

Stationary array

Element signals are aligned through delays/phases; pointing changes electronically.

Schematic silhouettes and functions of three families, without a common scale of sizes or beam patterns. The array is not represented as a filled metal aperture. LOFAR provides a construction example of stationary polarization elements. ASTRON

A stationary array can change its sensitive direction electronically by aligning element phases or delays. Its antennas do not turn like a large dish. Their individual patterns, the horizon and sidelobes still matter. LOFAR, for example, uses crossed dipoles for two polarizations; its LBA includes an LNA in the antenna head, while HBA elements are also combined at tile level. These are specific solutions, not a recipe for every array. ASTRON: LOFAR antenna construction.

Polarization describes the electric field’s behaviour. A linear receiving element responds to the field’s projection: rotating it relative to a fully linearly polarized wave changes the received power. An ideal element’s relative response follows cos²α:0.75 at 30° and 0.25 at 60°. Two orthogonal channels can reveal more about the field; they are not automatically two independent copies of every polarized source.

The reflector must preserve geometry and phase alignment

A distant source produces an approximately plane wave at the antenna. A paraboloid directs parallel axial rays to its focus with equal path lengths from a common entrance plane. Its shape therefore does more than make drawn lines intersect: it preserves phase alignment among contributions from different parts of the aperture. The article on telescope capabilities explains the connection between aperture and detail.

A paraboloid preserves equal path length to the focus

A paraboloid preserves equal path length to the focusz = 30 mFf = 10 mD = 25 m
Our exact geometric section: D=25 m, f=10 m, z=r²/(4f). Both axes use 12 px/m. Six rays at r=±3/7/11 m travel from z=30 m to reflector and focus F, each a 40 m path. Directions follow reflection, not merely connected points. Feed shape, supports and base are schematic; their blockage, diffraction, fields and antenna pattern are not computed. Not a manufacturing drawing or physical test. NRAO

Our teaching section has diameter D=25 m, focal length f=10 m and an entrance plane 30 m above the vertex. The surface follows z=r²/(4 f), where r is distance from the axis. At the edge, r=12.5 m gives z=3.90625 m. Each displayed ray travels 40 m in total to the focus. This is our geometric model, not a construction drawing for a real 25 m antenna; diffraction, horn shape and support blockage are not calculated. NRAO: deriving the paraboloid.

Manufacturing begins with requirements for band and accuracy. A backing structure supports panels or conducting mesh; surface positions are checked against the specified geometry. Mesh can suit longer wavelengths, but its cell size, conductivity, connections and shape do not become correct simply because it looks metallic. Shorter wavelengths require smaller geometrical errors. No universal mesh opening suits every frequency.

Making accurate individual panels is not enough: their positions on the supports must form one consistent surface. Radio holography is one method: compare received field amplitude and phase with a reference channel, derive a deviation map and adjust the surface. A source at finite distance requires near-field corrections. The authors of the 2006 evaluation of two ALMA prototypes reported setting their reflector surfaces to 17 μm RMS. That is their measured prototype result, not a property of our assumed dish. Radio holography method, prototype team’s result.

The feed must be placed and oriented correctly; supports and enclosures may block the aperture or couple unwanted radiation from warm ground into the receiver. Checks then cover geometry at different elevations, pointing and the sensitivity pattern. Gravity, solar heating and wind can alter the shape after assembly. A smooth-looking dish does not prove a correct beam pattern.

Branch · Wave and apertureOptional · you may skip this

Deeper: wavelength, angular scale and surface errors

For frequency ν in hertz and diameter D in metres:

λ = (c)/(ν), θ_(scale) = (λ)/(D).

c=299 792 458 m/s. At 1420 MHz, λ≈0.211121 m; with D=25 m, θscale≈0.008445 rad≈29.03 arcmin, about half a degree. At 20 MHz the wavelength is about 14.99 m. Longer wavelength at fixed D gives a broader characteristic angular scale. We do not call λ/D a universal half-power beamwidth: its coefficient and sidelobes depend on aperture illumination, blockage and the definition of width.

Radio waves do not make a reflector insensitive to errors. For a statistical model of random surface errors with RMS σ, the Ruze estimate for the directed coherent contribution is:

η_(s) = exp⁡ ⁣[ − (((4πσ)/(λ)))^(2)].

σ describes surface error, not an already doubled wavefront error. The 4 π factor comes from an approximately doubled path change on reflection. With σ=2 mm and our λ≈0.211121 m, ηs≈0.9859; at λ=3 mm the same statistical model gives about 3.31×10⁻³¹. This is loss of the aligned directional contribution, not a claim that metal absorbed all the energy. Systematic deformation, correlated errors and actual sidelobe structure need separate analysis. NRAO: assumptions for surface accuracy.

An amplifier raises signal and noise together, adding noise of its own

A very weak antenna voltage must reach the electronics. A low-noise amplifier, or LNA, raises its level and helps make later stages’ noise less important when referred to the input. The amplifier uses supply energy: a weak input controls a stronger output. It does not automatically separate the cosmic signal from everything else; it amplifies interference too and adds its own noise.

Loss before the first amplification, input matching, power supplies, shielding, temperature and operating band therefore matter. A long cable ahead of the LNA can attenuate the useful contribution and add thermal noise. Moving amplification closer to the antenna is often useful, but strong interference can overload an amplifier. Preselection may then be needed, and its losses also enter the budget. There is no universal rule that every filter must follow the LNA.

Noise temperature expresses noise power referred to a defined point in the chain. It need not equal the physical temperature of metal or a transistor. Sky, atmosphere, warm ground entering sidelobes, losses and electronics contribute differently. ALMA, for example, uses cryogenic receiver assemblies: a solution for particular frequencies and technologies, not evidence that every radio receiver needs the same cryostat. ALMA: receivers and cooling.

Branch · Noise chainOptional · you may skip this

Deeper: the same cable before and after amplification

Consider an explicitly assumed matched chain near 1420 MHz. Antenna contribution TA=50 K; amplifier equivalent input noise Tr=30 K; linear power gain G=100. The cable has loss ratio L=2 and physical temperature Tp=290 K. These are teaching assumptions, not an instrument specification or our own measurement. Decibels cannot replace linear G and L in these expressions.

Temperature referred to the antenna terminals is:

T_(no loss) = T_(A) + T_(r),; T_(before) = T_(A) + (L − 1)T_(p) + LT_(r),; T_(after) = T_(A) + T_(r) + ((L − 1)T_(p))/(G).

Without cable loss the result is 80 K. A cable before the amplifier gives 400 K; after it,82.9 K. In the last case, subsequent receiver noise is assumed zero; a nonzero contribution would also need referring through preceding gains. This follows from the order of gain and loss, not an amplifier somehow cooling the cable. NASA/JPL: cascaded noise and reference planes, equations 7.13–7.20.

In the classical approximation for this band, available noise power in one matched channel is P=kBTsys, where k=1.380649×10⁻²³ J/K and B is noise bandwidth in hertz. Output power after amplification differs; temperature and signal must be compared at the same point. Our 80 K cannot be transferred mechanically to a millimetre receiver or a different sky.

Frequency can be shifted; recorded numbers need a reference scale

A common approach mixes the radio signal with a local oscillator, or LO. The nonlinear mixer produces sum and difference frequencies. A filter selects an intermediate band that is easier to amplify and digitize. The oscillator is a separate mixer input; the celestial signal does not pass through the generator as a stage in series.

From field to calibrated numbers: quantities and units change

  1. 1. Antenna and feedAnalog voltage · V · RF 1419.5–1420.5 MHz
  2. 2. Low-loss RF preselectionVoltage · V · selected RF band
  3. 3. Low-noise amplification, LNAVoltage · V · amplified RF band
  4. 4. MixerVoltage · V · difference and sum bands
  5. 5. IF and antialias filteringVoltage · V · IF 19.5–20.5 MHz
  6. 6. ADCNumerical samples · ADU · 64 MS/s
  7. 7. FFT and squared magnitudeSpectral powers · relative ADU²
  8. 8. Averaging and calibrationEquivalent contribution ΔTₐ · K
  9. 9. Data recordingTime, frequency, pointing and settings

Assumed chain: Tₐ=50 K, Tᵣ=30 K, G=100, L=2, Tₚ=290 K

Chain arrangementTsys / K
No additional loss80
Loss before LNA400
Loss after LNA82.9
A teaching sequence of quantities, not a circuit schematic or instrument specification. The 1420 MHz RF centre is assumed, not the exact HI rest frequency. A real ADC takes IF in the first Nyquist zone 0–32 MHz. LO and clock are separate inputs; power and service networks are omitted. Relative FFT powers need calibration: ADU² are not watts. The three noise cases below are separately specified matched models with no additional noise from later stages. NRAO

In the assumed example RF=1420 MHz, LO=1400 MHz and the selected IF=20 MHz. Megahertz describe oscillation frequency; volts describe voltage amplitude. Those units are not interchangeable. The frequency reference sets the oscillator scale; the ADC clock sets sampling instants. Their accuracy, stability and known relationships support correct frequencies, phases and time stamps. A particular receiver may use different conversions or direct digitization.

An ADC converts analog voltage into discrete numerical samples. It needs a defined input band, sample rate and sufficient amplitude range. Strong interference can saturate an analog stage or clip samples; weaker interference changes the data even without obvious overflow. Increasing display colour contrast does not fix this physically.

Branch · Frequency conversionOptional · you may skip this

Deeper: difference frequency, image band and aliasing

For two unit-amplitude harmonics, a mathematical multiplier gives:

cos⁡(2πν_(RF)t)cos⁡(2πν_(LO)t); = (1)/(2)cos⁡ ⁣[2π(ν_(RF) − ν_(LO))t]; + (1)/(2)cos⁡ ⁣[2π(ν_(RF) + ν_(LO))t].

Here the products are 20 and 2820 MHz; the output filter keeps the first band. But RF=1380 MHz with the same LO also produces a 20 MHz difference. This image band requires input selectivity or another sideband-separation method. An IF peak alone cannot uniquely determine the original frequency without knowing the chain. NRAO: superheterodyne reception.

Suppose the selected IF is 19.5–20.5 MHz and a real-valued ADC samples at fs=64 million samples per second. In our ordinary first-Nyquist-zone scheme this lies within 0–32 MHz; unwanted high-frequency components must be suppressed before the ADC. A 44 MHz cosine at this fs, for example, produces the same sample sequence as a 20 MHz cosine. A good-looking graph cannot restore that lost information.

For an FFT of N=6400 samples, frequency spacing fs/N=10 kHz and block duration N/fs=0.0001 s. That is not the entire observation time: spectra can be averaged further. Windowing changes the response and equivalent noise bandwidth, so grid spacing is not always B in the radiometer calculation. Bandpass sampling in other zones is possible but is not assumed here. NRAO: DFT, sampling and power spectra.

A spectrum, time series and image answer different questions

Raw samples describe voltage over time. Measuring power involves squared amplitude and averaging; simply averaging positive and negative voltages is a different operation. A spectrometer examines contributions at different frequencies. After an FFT, accumulating squared magnitudes of complex coefficients produces a power spectrum. Saving only these powers no longer preserves the original phases completely.

A time series in a selected band shows variation, such as pulses or bursts. A spectrum shows frequency dependence. A spectrogram combines time and frequency, with colour representing a specified quantity and scale. A sky image also needs directional information: beam scanning, multiple beams or correctly combined array measurements. One dish recording one power value does not immediately supply a full-frame photograph.

ADC numbers may initially be in arbitrary digital units, ADU, and their squares in ADU². Conversion to volts, watts, equivalent antenna temperature or flux density requires calibration. A known noise contribution, reference load or source helps estimate response, but does not remove its frequency and time dependence. Background and source are checked separately, and settings and time stamps recorded.

A gas line calls for spectral position and profile; pulsars also require timing; polarization needs relationships between field channels. The NRAO 20-metre telescope’s own documentation assigns different bands and channel counts even to different spectrometer modes. One instrument’s settings are not a universal recipe for every SDR. NRAO 20 m: spectral modes and on/off.

Averaging reduces uncertainty without making noise power vanish

Random voltage may have a mean near zero, but its square is positive. Noise power therefore remains as a baseline. Longer observing estimates the mean more accurately when conditions remain sufficiently stable and the measurements provide new independent information. A weak added source contribution can be distinguished from fluctuations of a large background without exceeding the entire noise power.

Ideal RMS falls with time while mean noise power remains

Conditions: Tsys=80 K, B=10 kHz, one polarization. Both axes logarithmic; τ is total time, without overheads.

Ideal RMS falls with time while mean noise power remains0.010.030.10.311101001000σT / Kτ / s
Total power onlyIndependent on/off, half the time each
Our calculation of white stationary noise. Black: σT=80/√(10000τ) K; dashed: independent on/off, each τ/2. The same total time gives twice the RMS for on/off. Gain drift, overheads, atmosphere, interference and confusion from unresolved sources are absent, so extending the curves does not guarantee actual sensitivity. This is uncertainty, not mean noise power falling to zero. NRAO

In the stated ideal case, quadrupling time halves random RMS. This does not promise unlimited improvement: gain drift, atmospheric changes, interference or unknown background response impose other limits. Samples after windowing and overlap are not all independent. Excessive averaging can also hide the short event the data were intended to capture.

Branch · Sensitivity limitOptional · you may skip this

Deeper: bandwidth, time, polarizations and the on/off cost

The ideal estimate for stable white noise is:

σ_(T) = (T_(sys))/(√(n_(pol)Bτ)).

Tsys and σT are in kelvins, B is equivalent noise bandwidth of one measured channel in hertz and τ is integration time in seconds. Our example uses npol=1; npol=2 can apply when correctly averaging two independent, equally sensitive polarization estimates of the same signal. This is not automatic for an arbitrary polarized source. For a spectral channel B is neither the entire SDR band nor its sample rate.

At Tsys=80 K, B=10 000 Hz and τ=10 s, σT≈0.253 K; at τ=40 s it is about 0.1265 K. These are our assumed-chain calculations, not a named product’s claimed sensitivity. Widening a channel changes more than noise: a narrow line can be diluted and frequency information lost. Bigger B is therefore not always better.

An independent check comes from statistics. A normalized Gaussian voltage V has mean square 1 and fourth moment 3, so the variance of V² is 3−1=2. For N_eff=2 Bτ equivalent independent real samples, the RMS of the mean square is √(2/N_eff)=1/√(Bτ). This is not the count of every raw ADC sample: adjacent samples of a narrowband signal can be correlated. Mean power does not become zero. NRAO: radiometry and its assumptions.

An on−off difference contains noise from both measurements. With the same background and Tsys, and independent errors:

σ_(on − off) = (T_(sys))/(√(n_(pol)B))√((1)/(τ_(on)) + (1)/(τ_(off))).

Splitting 40 s total into 20 s on and 20 s off gives about 0.253 K, compared with 0.1265 K for all 40 s on source without a separate off. Differencing can control background and drift better, but does not remove random noise for free. Actual switching also has overheads and its own conditions. NRAO: limitations and differential reception.

A peak's origin is checked in several independent ways

This teaching spectrum has two contributions: a broader assumed celestial line and a narrow injected RFI feature. We know their origins because we specified the analytic functions. It is not an observation, an FFT of a real recording or proof that a broad peak must be celestial and a narrow one artificial.

The component origins are specified, not inferred from width

The component origins are specified, not inferred from width02468-0.3-0.1500.150.3ΔTₐ / KνRF − 1420 MHz
Injected celestial lineInjected RFIComponent sum
Our analytic injections, not observations or an FFT: a 1.5 K Gaussian line with σ=0.04 MHz at offset 0; 6 K RFI with σ=0.008 MHz at offset +0.18 MHz. Origins are known by construction. The y axis is an assumed ΔTₐ contribution after constant-baseline removal, not gas temperature. 201 points at 0.003 MHz spacing; no real channel response or random noise. Width alone does not classify an origin. NRAO

Actual celestial lines can be narrow and interference broad or variable. Harmonics, an unwanted sideband, local oscillators, power-supply pulses and saturation can also produce structure. Weak interference is dangerous not only because it masks everything: it can add a plausible false detail. NRAO: RFI sources and consequences.

Checks must match the hypothesis. Change direction or wait for the source’s predicted transit; compare background; examine repeat recordings and an independent instrument. Check behaviour when the tuning changes, using the known LO/IF arrangement: a local artifact and a celestial line need not move alike. Monitor overload and local electronics too. No single test is a universal detector of extraterrestrial origin.

Disconnecting an antenna can separate some receiver problems from external contributions, but the sky is not the only thing removed: external interference also arrives through the antenna, and an unmatched open input changes conditions. This is a chain check under its documentation, not proof that a peak was cosmic. Comparisons also require a proper load and unchanged remaining settings.

A radio-quiet site, low-noise power supplies, shielding and selectivity reduce problems before recording. Processing can flag or discard contaminated intervals, but a saturated chain and lost information cannot be restored by an attractive colour palette. Brighter colour is not equivalent to greater scientific reliability.

The same kind of wave supports different scientific measurements

Radio data probe gas through spectral lines, pulsar timing, solar and planetary bursts, synchrotron emission from charged particles in magnetic fields, and galaxy structure. This is not a search for broadcasts from every star. Band, polarizations, time resolution, beam and calibration are selected for the task. NASA: the radio sky. NRAO: synchrotron emission.

Many bands are observed from the ground; daylight is not the same obstacle it is for faint optical stars. The atmosphere, however, is neither transparent nor stable at every frequency. The ionosphere changes propagation of some low-frequency radiation, while water vapour matters at millimetre wavelengths. Radio is not universally independent of weather. NOAA: the ionosphere.

A space receiver makes sense when the desired band is inaccessible or distorted on Earth, or a different measurement environment is needed. NASA Wind carries WAVES, with radio reception and electric/magnetic-field recording. A radio telescope in space therefore need not resemble a large optical-style dish. NASA: Wind instruments.

To explain a result, follow the whole chain: which fields and directions the antenna accepts; how noise and bandwidth change; how frequency is converted and timing defined; which quantities are stored; how calibration works; and which independent checks separate sky, hardware and interference. A peak then becomes a measurement with known conditions, rather than merely an interesting shape on a screen.