H2-1 · What it is (and what it is NOT): scope boundary & typical deployments

Goal: define “microwave/mmWave backhaul” in one screen, set hard boundaries, then map the common deployment shapes.

Microwave / mmWave backhaul is the wireless transport hop that carries Ethernet or SDH traffic between a cell site (or remote node) and the aggregation / metro network. It sits between “base station access” and “core transport,” turning framed digital traffic into a stable RF link with the availability and spectral compliance required for carrier service.

This page focuses on the radio transport chain (RF conversion, IQ modulation/demodulation, low phase-noise LO/PLL, and interface-level Ethernet/SDH integration). It does not cover DU/CU compute, O-RAN RU transceiver/DPD specifics, optical transport (DWDM/ROADM/OTN), or timing-switch (PTP/SyncE) architecture details.

Microwave vs mmWave (E-band) — practical engineering boundaries:

• Distance vs throughput: higher carrier frequency enables wide channels and high throughput, but pushes tighter requirements on alignment, phase noise (EVM), and fade margin.
• Weather sensitivity: mmWave links typically demand a more deliberate plan for rain fade and availability (fade margin + adaptive modulation/coding policy), especially for strict SLAs.
• Mechanical realities: narrower beams at higher bands raise sensitivity to antenna pointing error and wind-induced drift; this directly shows up as SNR/EVM degradation and ACM downshifts.
• Spectral compliance: as channels widen and QAM order increases, meeting spectral mask / spurious constraints depends heavily on LO purity, mixer spurs, and PA linearity.

Typical deployment shapes:

• PtP (point-to-point): the dominant mode for high-capacity site-to-aggregation hops; predictable interference and throughput.
• PtMP (point-to-multipoint): useful for multi-site access; shared medium means scheduling and load can drive throughput variance.
• 1+1 protection: redundancy to maintain availability under fades or hardware faults; commonly paired with conservative ACM thresholds.

Reader expectation set: the rest of this page explains how RF conversion, LO/PLL quality, and IQ impairments map into “what the network sees” (throughput stability, EVM/BER margin, and link availability).

H2-2 · Reference architecture: IDU/ODU split and the signal path end-to-end

Goal: lock the end-to-end “golden path” (interfaces → baseband → RF → antenna), then show the feedback loops that keep EVM/BER stable outdoors.

A practical way to reason about backhaul radios is to separate the design into an Indoor Unit (IDU) and an Outdoor Unit (ODU). The split is not only mechanical; it clarifies where digital framing ends and where RF impairments begin, which helps diagnose field issues without guessing.

IDU typically contains:

• Ethernet / SDH interfaces (PHY/framer, buffering, alarms/counters) to present a transport endpoint.
• Baseband modem functions (mapping, modulation/demodulation control, and adaptive modulation/coding policy) to keep throughput stable under fades.
• Management plane (telemetry, configuration, and event logs) to expose “why the link stepped down” rather than only “that it stepped down.”

ODU typically contains:

• RF/IF conversion chain (mixers, IF filtering, gain stages) that sets spurs/image rejection and noise figure.
• LO/PLL synthesizer and distribution where phase noise and spurs often become the hidden EVM limiters for high-order QAM.
• PA/LNA + antenna / waveguide that defines EIRP, receiver sensitivity, and thermal-limited operating regions.

The most important “systems insight” is that a backhaul radio is not a one-way signal chain. It stays within spec by closing a few lightweight loops that continuously fight outdoor drift:

• AGC loop: observes IF/baseband level (and sometimes SNR proxies) → adjusts analog/digital gain to protect dynamic range and EVM.
• Tx power / linearity loop: observes EVM/adjacent leakage proxies + temperature → trims output power or bias to stay within spectral mask.
• Frequency / lock loop: monitors lock detect and residual frequency error → maintains carrier/LO integrity to avoid demod stress and throughput collapse.
• Calibration loop: tracks temperature and I/Q imbalance drift → updates correction coefficients so the constellation remains “round” over time.

Boundary reminder: interface timing (PTP/SyncE) may exist around these blocks, but detailed timing-switch architecture belongs to the “Timing Switch” sibling page.

H2-3 · Bands, duplexing, and channelization: what drives hardware choices

Goal: turn “band / duplex / bandwidth” choices into a clear hardware cost chain (phase noise, linearity, filtering, thermal, alignment).

Band selection, duplexing method, and channel bandwidth are not “RF preferences.” They set the constraints that ultimately decide whether the link can hold a stable throughput under fades while staying inside spectral mask and EVM limits. The practical chain is: throughput / distance / availability → SNR & EVM margin → phase noise + linearity + filtering + thermal.

Band choice directly shifts the hardware difficulty:

• PA output and efficiency: higher bands often reduce the “easy” linear output power. The system-level symptom is earlier ACM downshifts and less headroom during rain fades.
• Phase-noise pressure: higher-order QAM and wide channels tighten EVM tolerance, making PLL/LO purity a top limiter instead of an afterthought.
• Packaging and antenna mechanics: higher bands push tighter routing, shielding, and connector/waveguide sensitivity; narrower beams raise alignment sensitivity and wind-induced drift impact.

Duplexing choice changes the dominant interference paths:

• FDD (duplexer-based): filtering and isolation become the bottleneck. Insufficient isolation shows up as Rx desense (weaker sensitivity, BER spikes under low SNR). Filter behavior can also stress spectral compliance and EVM margin.
• TDD (switch-based): fast switching creates transient risks. Poor gating/isolation can cause Tx→Rx leakage during turn-around, leading to burst errors or unstable demod performance even when average SNR looks acceptable.

Channel bandwidth and modulation order tighten both LO and PA requirements:

• Bandwidth ↑ increases the chance that spurs/images land near the wanted channel; filters must leave margin for skirts and drift.
• QAM order ↑ reduces constellation spacing, so the same phase noise or nonlinearity consumes more of the EVM budget.
• System outcome: passing spectral mask + EVM often forces either cleaner LO, more linear PA operation (lower output or better biasing), or stronger filtering—each with power/thermal cost.

Regulatory note (kept brief by scope): spectrum planning constrains usable channels and EIRP limits, which become fixed inputs to frequency planning and PA/antenna headroom.

H2-4 · Up/down conversion fundamentals: mixers, images, LO feedthrough, filtering

Goal: build a frequency-planning mindset—identify images, LO leakage, and spurs early, then assign filtering and linearity budget by stage.

A backhaul radio moves a complex-modulated waveform from baseband into RF and back again using one or more conversion steps. Implementation details vary, but the engineering risks repeat: image landing, LO feedthrough, and spurs/intermod—all of which can break spectral mask compliance or consume EVM margin.

Typical conversion paths (kept at system level):

• IF architecture: Baseband (I/Q) → IF → RF. Easier to place filters for image/spur suppression, but adds frequency-planning complexity.
• Direct conversion: Baseband (I/Q) → RF. Short chain and integration-friendly, but more sensitive to DC offset and LO leakage shaping the spectrum.

How to budget gain, noise, and linearity without drowning in formulas:

• Near the antenna: prioritize noise figure and overload resilience (avoid compression/desense under strong signals).
• In conversion stages: prioritize spur hygiene (keep LO-related products out of the wanted band and leave filter margin for drift).
• Near the PA / output: prioritize linearity headroom to meet spectral mask while keeping EVM stable at the required throughput.
• Validation habit: sweep frequency, output power, and temperature to confirm that spurs do not “walk into” the channel under real conditions.

The conversion chain is best treated as a frequency-planning exercise first, and a component-selection exercise second. If planning is wrong, no single “better mixer” can fully rescue the spectrum.

H2-5 · IQ mod/demod and impairments: EVM as the north-star metric

Goal: connect IQ impairments to a single system outcome (EVM → BER margin → ACM mode → throughput stability), then map the six most common EVM killers.

In a microwave/mmWave backhaul radio, IQ modulation/demodulation is the step that places complex vector information (amplitude and phase) onto a carrier and recovers it on the receive side. For high-order QAM and wide channels, EVM (Error Vector Magnitude) is the most practical north-star metric because it captures the combined impact of IQ imbalance, LO/PLL quality, residual frequency error, and nonlinearity.

The system chain is simple and useful in the field: EVM margin determines BER margin, which drives ACM downshift frequency, which decides whether throughput is stable or “steps down” under fades.

Six common EVM killers (symptom → likely cause → practical check):

• I/Q gain mismatch: constellation stretches into an ellipse → I/Q amplitudes differ → compare EVM shape across frequency and temperature; calibration should reduce axis-dependent spread.
• Quadrature phase error: points skew with cross-axis coupling → I/Q not at 90° → check whether distortion is frequency-dependent; phase calibration should straighten the pattern.
• DC offset / carrier leakage: constellation shifts off-center or a fixed tone appears → imperfect balance or DC bias → verify “quiet” spectrum for a persistent spur; DC correction should reduce it.
• Residual carrier frequency offset (CFO): constellation slowly rotates → LO frequency error or marginal lock → correlate rotation rate with lock status and temperature drift.
• Phase noise: constellation becomes “fuzzy” (random angular jitter) → LO/PLL noise dominates → compare EVM across modulation orders and channel widths; higher order/wider channels amplify sensitivity.
• PA nonlinearity / compression: EVM worsens sharply with output power and spectral mask tightens → AM-AM/AM-PM distortion → run a power sweep to locate the linear operating window.

Boundary note: deeper baseband equalization and OFDM receiver algorithms are intentionally out of scope here; the focus is the radio-level impairment map and calibration mindset.

H2-6 · Low phase-noise PLL & LO distribution: from L(f) to EVM

Goal: make phase noise operational—identify loop “knobs” (benefit / side-effect / validation) and connect L(f) shape and spurs to EVM and spectral mask risk.

For wide channels and high-order QAM, phase noise often becomes the hidden EVM limiter. Even when received power looks adequate, a noisy LO spreads phase error across the constellation, reducing BER margin and triggering ACM downshifts. The right way to work this problem is to treat the LO as a system resource: PLL architecture + loop shaping + spur management + distribution integrity.

Common PLL/LO architecture choices (engineering-level):

• Integer-N vs Fractional-N: Integer-N tends to be spur-friendly but less flexible; Fractional-N expands frequency agility but demands stricter spur hygiene.
• Reference strategy: reference quality and distribution cleanliness can dominate close-in regions after loop shaping.
• LO distribution: buffers/splitters/isolation matter because any added noise or coupling propagates into every mixer, impacting both EVM and spurious emissions.

Practical “knob list” (benefit → side-effect → validation):

• Loop bandwidth: reshapes ref vs VCO dominance → too wide imports reference noise; too narrow leaves VCO drift → validate by L(f) crossover position plus lock/temperature behavior.
• Reference noise: improves the region where ref dominates → distribution coupling can erase gains → validate by swapping reference/distribution conditions and correlating EVM change.
• VCO noise: improves far-out regions that wide channels integrate → may raise power/thermal demands → validate with temperature sweeps and modulation-order sensitivity.
• Divider/PFD/CP noise: reduces mid-band “humps” → can increase spur sensitivity → validate by searching for platform-like bumps vs discrete spurs.
• Spur management: prevents discrete tones from landing near the channel or skirt → may constrain frequency plans → validate by sweeping channel plans and ensuring spurs stay away from skirts.
• Distribution isolation: limits LO feedthrough and cross-coupling → adds insertion loss and gain staging needs → validate by changing load/enable states and checking spur/EVM stability.

What to “read” in a phase-noise plot: the curve shape (which region dominates), the crossover (loop shaping), and spurs (discrete mask/EVM risks).