Inside a Spectrum Analyzer: Board Design and Shielding
A spectrum analyser is an instrument that measures amplitude against frequency, and it does so over an enormous dynamic range. It has to resolve a signal a hundred decibels below the strongest one present, sweep or capture across decades of frequency, and maintain its calibration through all of it. The board inside the instrument is where that performance is realised, and the design is dominated by shielding, grounding and the reference rather than by any single component.
What the Instrument Has to Do
The signal path begins at an input connector, passes through attenuation and possibly amplification, is mixed with a local oscillator to bring it to a fixed intermediate frequency, is filtered to a defined resolution bandwidth, and is finally detected and digitised. Each of those stages has its own requirement for linearity, for loss and for isolation between the signal and the local oscillator.
Modern instruments digitise earlier and do more of the work in the digital domain, which shifts the burden from analogue filtering to digital processing, but it does not remove the analogue front end. The converter has a finite dynamic range, so the analogue path still has to deliver the signal at the right level and with the interferers attenuated before conversion.
Shielding and Isolation
Dynamic range is limited by leakage rather than by noise in a well designed instrument. If a fraction of the local oscillator or of a strong input signal reaches the wrong part of the chain, it produces a spurious response that the instrument reports as a real signal, and no amount of digital processing can separate it afterwards. That is why the radio frequency sections are built in shielded compartments with the board divided by walls and the connections made through feedthroughs.
The board design supports that structure. Each compartment needs a continuous ground plane that is bonded to the walls along the whole perimeter, with vias close enough together that the seam between them behaves as a solid conductor rather than as a slot. Separating the local oscillator from the input path, and keeping the chains physically apart, is worth more than any filter that could be added later. Our component tolerance and reliability notes describe how those assemblies are assessed.
Reference, Calibration and Accuracy
An instrument’s accuracy depends on its timebase and on its reference level. The timebase sets the frequency axis, so its stability determines whether a measurement is repeatable from one day to the next, and it is usually locked to an internal oven controlled oscillator or to an external reference. The reference level sets the amplitude axis, and it is calibrated against a known source and stored as a correction table.
Those calibration constants live on the board in non-volatile memory, and the design has to make them recoverable. They are generated during manufacture with equipment that the user does not have, so losing them turns the instrument into a paperweight. Storing them in more than one place, and verifying them at start-up, is a small cost for a large risk reduction.
Layout of the Signal Chain
Each stage of the chain is laid out as a fifty ohm transmission line with a continuous reference plane, and the transitions between stages are the places where performance is lost. A mixer’s ports have to be matched, a filter’s input and output impedances have to match the chain and an amplifier with poor reverse isolation allows the local oscillator to leak back toward the input.
The power supply to the radio frequency amplifiers receives the same attention. A switching supply that powers the digital section is not connected directly to a low noise amplifier, and the analogue supply is derived linearly and filtered close to the load. Where a stage has its own shielding compartment, its supply and control lines enter through feedthrough capacitors so that the compartment remains a barrier at high frequency as well as at low frequency. Our design release checklist places those checks in the review sequence.
Digital Section and Interference
The digital section processes the digitised data at high speed, and its clock harmonics extend well into the frequency range the instrument is measuring. Keeping those harmonics out of the analogue path is a placement problem before it is a filtering problem: the digital circuits go on their own area of the board, their clocks are routed away from the analogue sections and their return currents are given a path that does not run beneath the sensitive circuitry.
Where the converter sits between the two domains, its layout determines the instrument’s noise floor. The converter needs a clean clock, a stable reference and a ground connection that does not carry digital return current, which usually means treating its ground as part of the analogue domain and connecting the two domains at a single point close to the converter.
Testing and Calibration in Production
Production testing verifies the amplitude accuracy, the frequency accuracy, the noise floor and the spurious responses across the whole band, and it is done against traceable references rather than against a second instrument of the same type. The calibration data generated by those measurements is written into the board’s memory, which means the test station and the board have to agree on a format that will survive a firmware update.
The most demanding measurement is the spurious response test, because it requires a clean source and a quiet environment. An instrument that is tested in a room full of switching supplies and radio transmitters will show spurs that come from the room rather than from itself, so the test area itself is part of the production equipment.
Mechanical Design of a Shielded Assembly
The compartments are formed by a metal frame that is soldered to the board or clamped to it, and the board’s contribution is a ground structure that the frame can bond to along its entire perimeter. That means a row of vias, or a continuous ground pad, under every wall, and it means the wall cannot cross a region where the ground plane has been removed for another purpose. Where the frame is soldered rather than clamped, the assembly becomes difficult to rework, which is a decision taken deliberately in instruments where the shielding is part of the calibration.
The connectors and feedthroughs that cross the walls are the weak points. Each one needs its own ground connection to the wall, and a connector that is grounded only through a trace behaves as a slot antenna at the frequencies the instrument is trying to measure. Keeping those transitions short, and giving each of them a dedicated ground path to the frame, preserves the isolation the compartments were built for.
Thermal Behaviour and Drift
An instrument that measures with a stated accuracy has to reach a stable temperature before it can be trusted, and the board is part of that. Amplifiers drift with temperature, the local oscillator drifts unless it is oven controlled, and the reference level shifts as the gain of the chain changes. Design measures reduce the drift by keeping the temperature stable rather than by making every part perfect.
That means separating the heat generating parts from the parts whose behaviour is temperature sensitive, and giving the sensitive sections a thermal path to the chassis rather than letting them sit in the warm air inside the instrument. Where a compartment contains a dissipating amplifier, the floor of that compartment should conduct into the chassis. Our thermal management article describes how those paths are arranged.
FAQ
Why are spectrum analyser boards built in shielded compartments? Because leakage limits the dynamic range. A fraction of one signal reaching another part of the chain produces a spurious response that cannot be removed later.
Why does the local oscillator need its own shielding? Because it is a strong signal at a known frequency, and any leakage into the input path appears as a spurious response at the frequency the instrument is measuring.
What happens if the calibration data is lost? The instrument loses its amplitude accuracy and cannot be restored without the manufacturer’s test equipment. Storing the data redundantly and verifying it at start-up is the standard mitigation.
