Multi-carrier microwave backhaul looks deceptively straightforward: add carriers, aggregate throughput, and let ACM keep the link honest. In practice, the phase noise budget is what quietly sets the ceiling on modulation order, XPIC performance, and how much capacity you can really bond before the radio starts “breathing” EVM. If you are architecting 2+0, 4+0, or wideband aggregated carriers (especially into the mmWave/E-band regime), phase noise stops being a component spec and becomes a system-level design discipline.
This post lays out a practical way to build a phase noise budget for multi-carrier links—what to allocate, what to measure, and where surprises usually appear. It’s written for the engineer who needs a link that closes on paper and survives production spread, temperature, ageing, and real adjacent-channel conditions.
Why multi-carrier makes phase noise harder (and more important)
In a single-carrier radio, phase noise mainly hurts you through:
- Common Phase Error (CPE): slow phase wander that rotates the constellation.
- Inter-Carrier Interference (ICI): for OFDM, close-in phase noise spreads energy into adjacent subcarriers.
- Reciprocal mixing: oscillator noise folds strong nearby interferers into your channel.
- Residual carrier/clock recovery stress: the DSP works harder, and performance becomes implementation-limited.
Multi-carrier aggregation amplifies the pain because the RF chain is now asked to maintain constellation integrity across multiple simultaneous channels. If you share a reference/LO across carriers, you can get correlated phase noise (sometimes helpful for cancellation), but you can also inject a single point of failure that degrades all carriers at once. If each carrier has its own synthesiser path, you trade correlation for an explosion in noise contributors and spurious interactions.
Industry direction is clearly towards bonding/aggregation of multiple carriers to keep scaling capacity, rather than relying on a single ever-wider channel. That approach is well established in microwave transport planning, where bonding provides near-100% utilisation and flexibility as you add carriers of different bandwidths or even different bands. The catch is that the phase noise penalty does not scale linearly with “number of carriers”; it often shows up as a step change when the worst offset region of the LO lands right where your demodulator is most sensitive.
Start with the requirements: your demodulator doesn’t care about dBc/Hz, it cares about EVM
The most reliable budgeting flow is to translate oscillator phase noise into an EVM contribution, because that is where modulation order, coding, and XPIC margins are actually decided.
A workable top-down method:
- Set EVM targets per modulation (e.g., 1024/2048/4096-QAM where applicable) with implementation margin. In backhaul radios pushing high spectral efficiency, that margin is often what keeps ACM from downshifting on hot days.
- Allocate an EVM slice to phase noise (for example 20–35% of the total EVM budget, depending on PA linearity, IQ imbalance, and channel conditions).
- Convert EVM allocation to integrated phase error (rms phase jitter) and then to a phase noise mask requirement over offset frequency.
For many QAM demodulators, a first-order approximation is that rms phase error (in radians) contributes approximately proportionally to EVM. The precise mapping depends on the receiver architecture and tracking loop bandwidth, but the key is this: close-in phase noise (typically 10 Hz–100 kHz offsets) is where high-order QAM gets strangled, whereas far-out phase noise is more likely to show up as adjacent channel leakage and reciprocal mixing issues.
Building a practical phase noise budget: what to allocate, and where it hides
A backhaul radio phase noise budget is best treated as a cascade of contributors referred to a common point (usually the RF carrier). Typical blocks:
- Reference oscillator (OCXO/TCXO/GPSDO/SyncE-derived): sets the low-offset “floor” your PLL can’t escape.
- Synthesiser/PLL residual noise: dominates mid-offset regions depending on loop bandwidth and divider ratios.
- VCO/DRO: dominates beyond loop bandwidth; in mmWave systems, multiplication pushes phase noise up by 20·log(N).
- Frequency multiplication and mmWave LO generation: often the single biggest “gotcha” moving from 18/23 GHz into E-band.
- Up/down-conversion mixers: add flicker upconversion and AM-to-PM effects if driven poorly.
- Clocking/jitter on data converters (if using high-speed DAC/ADC with digital up/down conversion): can appear indistinguishable from phase noise at RF.
Two practical rules that save time:
- Budget in offset regions, not as a single number. Treat 10–100 Hz, 100 Hz–10 kHz, 10 kHz–1 MHz, and 1–10 MHz as separate “bins” because different physics dominates each.
- Budget per carrier, then check the composite. A synthesiser that is “fine” for one carrier can break multi-carrier EVM when shared spurs or correlated noise land in a sensitive region for the aggregation DSP.
Standards reality check: spectrum masks and aggregated emissions are tightening
In regulated microwave bands, your phase noise plan cannot be separated from spectrum compliance. ETSI’s point-to-point radio standards continue to evolve, and the latest releases reinforce a theme system architects must internalise: each aggregated channel emission must meet its own mask even under aggregation. That matters because phase noise skirts and spurious structures can change when you aggregate carriers, share LOs, or alter loop bandwidths.
In other words, an aggregation mode that “works” in the lab may fail compliance in production if you haven’t treated phase noise and spurs as multi-mode behaviours, not single-mode plots.
Multi-carrier pitfalls: where phase noise bites in 2+0, XPIC, and carrier aggregation
Three failure modes show up repeatedly in microwave backhaul architectures:
1) XPIC and polarisation purity become phase-noise sensitive
XPIC is fundamentally a cancellation problem. If phase noise reduces the coherence between the desired polarisation and the interference estimate, the canceller leaves residue. The symptom is that the radio “has XPIC” but never quite reaches the promised throughput at high modulation—particularly when temperature or ageing shifts loop dynamics. A robust phase noise budget should explicitly allocate extra margin when XPIC is a requirement, not an optional feature.
2) Shared LO correlation: sometimes helpful, sometimes disastrous
Sharing a low-noise reference and LO chain across carriers can create correlated phase noise, which certain receiver structures track more effectively (CPE cancels similarly across carriers). However, shared spurs and fractional-N boundary effects can also line up across carriers, producing a repeatable but catastrophic EVM spike at specific channel plans. The engineering stance here is simple: design for correlation, but verify worst-case spur alignment across every aggregation combination you intend to ship.
3) Reciprocal mixing with real interferers, not test benches
Backhaul deployments often see strong adjacent links, co-sited radios, and dense band plans. Phase noise at offsets equal to the interferer spacing sets how much that interferer is “mixed” into your channel. This is one reason why updated radio-noise modelling guidance (for instance, ITU-R material covering background noise up to 100 GHz) matters: when you move higher in frequency and start relying on narrower fade margins, your receiver impairment stack-up has less room for sloppy oscillator skirts.
Recent oscillator and synthesiser direction: what’s actually improving
Two trends are worth factoring into 2024–2025 roadmaps:
- PLL residual noise floors and loop bandwidths are being pushed. Modern synthesisers can deliver very low phase noise at practical offsets with faster switching—useful for radios that need agile channel plans or rapid hitless switching. Technical literature from test-and-measurement vendors shows mainstream architectures still dominated by indirect PLLs, but with continued work on lowering residual floors to support wider loop bandwidths and better close-in performance.
- High-Q resonators remain the “endgame” for the lowest noise. Sapphire-loaded cavity oscillators and similar high-Q approaches demonstrate extremely low phase noise at microwave offsets. You may not put a lab-grade resonator into every ODU, but the direction of travel is clear: where ultra-high-order QAM and tight channel packing are commercial differentiators, oscillator quality is once again a product feature, not an afterthought.
The practical takeaway for architects: expect synthesiser datasheets to look better, but don’t assume your system will. Packaging, supply noise, digital coupling, and multiplication will happily erase paper gains unless you design the whole LO chain as a first-class subsystem.
How Novocomms Space helps: phase noise as a design-for-deployment problem
At Novocomms Space, we’re often brought in when teams have an uncomfortable gap between “meets spec in one mode” and “ships reliably across band plans, temperatures, and aggregated carriers”. Typical use-cases include:
- Low phase noise frequency conversion and LO chain design for microwave and mmWave payloads, where multiplication and reference distribution dominate the final phase noise profile.
- Architecture trade studies (shared LO vs per-carrier synthesis; analogue vs mixed-signal upconversion) tied to EVM and compliance, not just oscillator plots.
- Test strategy development: defining production-friendly measurements that correlate with in-field EVM and adjacent-channel performance, rather than relying on a single “phase noise at 10 kHz” number.
Whether your platform ends up in terrestrial microwave backhaul, satcom ground terminals, or dual-use links where spectral discipline matters, the same engineering truth applies: a phase noise budget that is traceable to EVM and validated across modes is the difference between a clever prototype and a product line.
Conclusion: treat phase noise like a shared resource, not a component attribute
A multi-carrier radio succeeds when every impairment has a sensible allocation, a measurement method, and a margin story. The phase noise budget is the thread that ties oscillator choice, synthesiser architecture, frequency plan, and DSP robustness into one accountable design. Build it in offset “bins”, tie it to EVM, and validate it across every aggregation mode and channel plan you intend to ship—because that is where the real failures hide.
If you’re pushing higher-order modulation, tighter channel packing, or more aggressive carrier aggregation and want a phase noise budget that stands up from lab to deployment, talk to Novocomms Space: https://novocomms.space/contact-us/.
