Spectrum Analyzer PCB: Design for a Low Noise Floor

The Board Sets the Noise Floor

A spectrum analyser measures the amplitude distribution of a signal across frequency, and what it can see is limited by its own noise floor. That floor is set by the receiver chain, and the receiver chain sits on a board. The chip selection, the filter, the low noise amplifier and the mixer all matter, but none of them can perform better than the interconnect that joins them.

This is the part of the design that receives the least attention and causes the most disappointment. A board with a split ground plane, a via stub in the RF path, or a GHz signal routed on standard FR-4 will raise the noise floor and degrade the measurement, and no amount of component quality compensates for it.

The reason is that a spectrum analyser board carries several incompatible signal types at once: very low level analogue RF, GHz frequency microwave transmission, and high speed digital processing for the display and control. Keeping those domains separated on one substrate is the central design problem.

What the Board Carries

• RF input and acquisition: the front end where a 50 ohm path has to be preserved from the connector inward.
• Attenuation and amplification: the low noise amplifier chain, where every unnecessary loss in the interconnect directly worsens the noise figure.
• Filtering and mixing: frequency selective elements and the conversion stage.
• Analogue to digital conversion and digital processing: the point where high speed digital noise enters the design and must not reach the analogue section.

The third and fourth items are where the difficulty concentrates. The converter and its digital interface are a significant noise source sitting in the same enclosure as a receiver whose sensitivity is measured in fractions of a microvolt.

Why This Is Not a Standard Board

• Frequency range from megahertz to gigahertz, so the design rules that apply at the top of the range govern the whole board.
• Extreme noise sensitivity, since the instrument’s purpose is to resolve small signals.
• Strict impedance control, typically 50 ohm single ended for RF paths and 75 ohm in specific video or broadcast applications.
• Susceptibility to EMI, crosstalk and return loss, all of which appear as measurement error rather than as an obvious fault.
• High frequency materials required for the RF sections, since standard FR-4 loss becomes significant above roughly 500 MHz to 1 GHz.

A single routing or via decision can raise the noise floor. That is the standard against which the layout has to be judged, and it is a much tighter standard than functional correctness.

Design Rules That Matter

• Calculate the RF trace impedance rather than estimating it. The geometry has to come from the actual stackup and the actual laminate, not from a previous project.
• Keep RF paths short and direct. Every millimetre of loss in front of the first amplifier adds directly to the system noise figure.
• Maintain a continuous reference plane beneath every RF trace. A split or a slot changes the impedance and the return path at the same time.
• Separate RF, analogue and digital regions physically, including on adjacent layers, and define where the grounds join.
• Design the power delivery network deliberately, since supply noise that reaches the RF chain appears as spurious content in the measurement.
• Treat the connector transition carefully. The path from the SMA or other RF connector onto the board is a discontinuity that has to be designed rather than drawn.

Verification follows from those rules: coupon based measurement of the manufactured impedance, and channel evaluation where the instrument performance depends on it. Coupon measurement is the practical form of TDR impedance testing in production, and it is the only way to know that the board matches the design intent.

Stackup by Instrument Class

• Four layers: hand held instruments where the RF structure has to stay economical. Workable, with limited isolation between domains.
• Six layers: mid range analysers, providing dedicated ground and power planes and better separation between RF and digital sections.
• Eight layers: laboratory grade equipment, where EMI behaviour and signal integrity justify the additional planes and the extra shielding layers.

The stackup decision determines how much isolation is available, and isolation is what keeps the digital section out of the analogue chain. Adding layers is usually cheaper than trying to solve a noise problem with filtering after the fact, because the noise path is physical rather than electrical.

Material Selection

• Rogers and comparable hydrocarbon ceramic laminates: the standard for RF sections, with low loss and stable dielectric properties.
• PTFE based materials: the lowest loss option, used where the noise floor requirement is most demanding.
• Hybrid stacks: high frequency material for the RF layers and FR-4 for the digital and control layers. This is the mainstream approach for cost effective instruments, and it is a construction to be qualified rather than assumed.

FR-4 remains usable for the low frequency sections of a hand held instrument, but not for the RF path. The dielectric loss of standard FR-4 above roughly 1 GHz is enough to degrade the measurement in a way that cannot be corrected downstream. Getting the stackup and material settled is the core of the fabrication review on this kind of board, and it is better done before the layout than after.

Grounding, Shielding and EMI Control

• A complete RF ground plane under the signal path, without cuts or slots.
• Via fencing along RF traces, a row of ground vias either side of the trace that ties the top and bottom ground planes together and confines the field. This is one of the highest value layout techniques on an RF board and costs nothing but via count.
• Shielding cans over the RF sections, which limit both emission and susceptibility.
• Separation of analogue and digital ground, joined at a defined point so the return currents do not share a path.
• Large copper pours to reduce coupling between adjacent structures and to provide a low impedance reference.

None of those is exotic, and all of them are cheap relative to the cost of a noise problem. The failure mode is that they are omitted because the layout looked acceptable on screen, and the consequence appears only in the measured noise floor.

Manufacturing Requirements

• Impedance tolerance within about plus or minus five percent, verified rather than assumed.
• Accurate layer to layer registration, since the RF structures depend on the alignment of the trace to its reference plane.
• Tight control of dielectric thickness, which together with trace width determines the impedance.
• Copper surface quality, because surface roughness contributes to loss at high frequency and varies between processes.
• Surface finish: ENIG or immersion silver is preferred on RF boards, for flatness, solderability and loss characteristics.

Those requirements are exactly the ones that separate a fabricator who builds high frequency boards from one who does not. A conventional shop can produce the same layer count and the same outline, and still fail on the measured impedance and the loss.

Assembly

• Precision placement of RF devices, including the amplifiers, filters and mixers where a small positional error changes the electrical behaviour.
• Controlled reflow profiles, matched to both the high frequency laminate and the RF devices. A profile developed for standard FR-4 assumptions may exceed the limits of the laminate.
• Optical and X-ray inspection after assembly, followed by RF performance measurement, because a board that passes continuity test can still fail on the measurement that matters.

The SMT assembly stage on an RF instrument is a specialised operation rather than a generic service, since both the process and the test regime differ from those used on digital boards.

Cost

• Four layer RF prototype on high frequency material: roughly 180 to 350 dollars.
• Six layer hybrid board with high frequency and FR-4: roughly 400 to 700 dollars.
• Eight layer board using high frequency material throughout: roughly 900 to 1,500 dollars.

Those are board costs; the assembled instrument board commonly exceeds 600 dollars and can go well beyond 2,000 depending on the RF component content. The cost drivers are the material, the layer count, the processing tolerance and the assembly and test content. The verification that goes with it is not optional either, and the testing regime on an RF instrument board typically includes measurement of the parameters the instrument specification depends on, not just functional pass or fail.

Common Design Errors

• A split ground plane under the RF routing, the single most damaging mistake.
• Using FR-4 for a GHz signal path and expecting the loss to be acceptable.
• RF traces that are longer than necessary, adding loss in front of the first amplifier where it costs the most.
• A stackup that does not support the impedance structures the design assumes.
• No shielding strategy, leaving the RF sections exposed to the digital noise generated on the same board.

Selecting a Manufacturer

• High frequency material processing capability across the common RF laminate families.
• Impedance test reporting, so the result is measured and documented.
• RF and microwave board experience, which is what predicts whether the process can hold the tolerances.
• One stop assembly capability, since the RF assembly and the RF test are part of the same competence.

Applications

Spectrum analyser boards appear in communications test equipment, wireless signal measurement instruments, hand held analysers, EMC and EMI test equipment and microwave laboratory instruments. All of them share the requirement that the board not contribute measurable error to the instrument’s own measurement.

Frequently Asked Questions

Can FR-4 be used for a spectrum analyser board? Only for the low frequency sections. The RF path requires a low loss high frequency material, since FR-4 loss above roughly 1 GHz degrades the measurement in a way that cannot be recovered later in the chain.

What impedance is used? 50 ohm is standard for RF paths, with 75 ohm appearing in specific video and broadcast applications.

How many layers are recommended? Six to eight for professional equipment, four for hand held instruments where the structure has to remain economical.

Why is shielding necessary? Because the instrument’s own digital processing generates noise next to a receiver that has to resolve very small signals. Shielding and ground structure are the mechanisms that keep the two domains apart.

What is the most common cause of a poor result? A ground plane discontinuity under the RF path, usually introduced during layout to make room for something else and never reconsidered.

Summary

A spectrum analyser board is an RF instrument on a substrate, and its measured performance is limited by the board as much as by the components. The receiver chain has to be kept short, the impedance calculated against the real stackup and maintained in production, and the reference plane kept continuous beneath every RF trace.

Material and stackup carry most of the responsibility. High frequency laminate for the RF sections, a hybrid construction for cost control where the design allows it, and a stackup with enough planes to separate RF, analogue and digital domains. Ground via fencing and shielding are cheap measures that prevent expensive problems.

Manufacturing then determines whether the design survives. Impedance verification within about five percent, accurate registration, controlled dielectric thickness and a finish suited to RF are the requirements, and they are the reason the work belongs with a supplier who builds RF boards as a normal part of their business rather than occasionally.