The trend of science and technology is changing rapidly.

How to Achieve Sub-1ps Clock Jitter in Multi-Radio IoT Devices: Power, Layout, and PLL Design Rules

Insights 250

Your dual-band Wi-Fi 6 + BLE 5.3 gateway works fine on the bench—but in the field, throughput drops by 40% when the cellular modem transmits. You checked antennas, filters, and coexistence protocols… but the real culprit? Clock jitter on the 40 MHz reference oscillator induced by LTE power amplifier (PA) switching noise.

In modern multi-radio systems, timing is everything. A jitter of just 2 ps RMS can degrade EVM (Error Vector Magnitude) by 3–5 dB—enough to drop from 256-QAM to 64-QAM, halving your data rate.

At ChipApex, we’ve debugged jitter issues in smart home hubs, industrial routers, and medical telemetry devices. In this guide, Senior FAE Mr. Hong reveals how to achieve <1 ps RMS integrated jitter (12 kHz–20 MHz)—even with a noisy switching regulator nearby.


Why Clock Jitter Matters More Than Ever

With higher-order modulation (1024-QAM), wider channels (160 MHz), and tighter timing budgets (TDD sync), jitter directly limits system performance:

StandardMax Allowable JitterImpact of Excess Jitter
Wi-Fi 6/6E~1.5 ps RMSEVM degradation → lower MCS
5G NR FR1<0.5 ps RMSOFDM symbol distortion → BLER ↑
BLE Long Range~5 ps RMSPacket loss in Coded PHY
Ethernet TSN<100 fs (!)Time synchronization error

🔬 Real case: A smart speaker failed FCC radiated emissions—not because of RF, but because crystal oscillator harmonics were amplitude-modulated by DC-DC ripple, creating spurs at 800 MHz.


The Three Sources of Clock Jitter (and How to Fix Them)

Source 1: Power Supply Noise → VCO Sensitivity

All oscillators have power supply rejection ratio (PSRR). But most CMOS XO datasheets only specify it at 100 Hz–10 kHz—not at MHz-range switching noise from buck converters.

✅ Solution:

  • Use an ultra-low-noise LDO (e.g., TPS7A47, LT3045) with >60 dB PSRR at 1 MHz
  • Add LC π-filter after LDO: 1 µH + 10 µF (ceramic) + 100 nF
  • Never power XO from a switching rail—even with a ferrite bead

⚠️ Critical: Measure actual ripple at XO VDD pin with 1 GHz scope + 10x probe. We’ve seen 50 mVpp ripple where simulation predicted <1 mV.


Source 2: RF Coupling → Substrate/Trace Crosstalk

When a 2.4 GHz PA switches at 100 W peak, its harmonics (e.g., 800 MHz, 1.2 GHz) can couple into:

  • Crystal tank circuit → frequency pulling
  • XO output buffer → amplitude modulation
  • MCU clock input → duty cycle distortion

✅ Solution:

  • Place XO ≥20 mm away from RF PAs, antennas, and high-current traces
  • Surround XO with solid ground guard ring (stitched to plane every 2 mm)
  • Use differential XO (LVDS/LVPECL) if jitter budget <0.5 ps—single-ended is more susceptible
  • Shield XO with metal can or grounded copper pour (if space allows)

📏 Rule: Crosstalk voltage ≈ M × di/dt. At 1 A/ns, even 1 pH mutual inductance = 1 mV coupling—enough to modulate a 1.8V XO.


Source 3: Poor PLL Configuration → Amplifying Noise

Many MCUs and SoCs integrate PLLs to multiply the reference clock. But if the PLL loop bandwidth is too wide, it passes XO phase noise; if too narrow, it amplifies internal VCO noise.

✅ Solution:

  • Set PLL loop bandwidth to 1/10th of reference frequency (e.g., 4 MHz for 40 MHz XO)
  • Enable fractional-N dithering only if needed—can increase jitter
  • Use external low-jitter XO instead of internal RC oscillator for critical radios
  • Verify integrated jitter in datasheet: look for 12 kHz–20 MHz range (not just “typical”)

💡 Pro tip: Some SoCs (e.g., NXP i.MX RT1170) offer dual PLL paths—use separate PLLs for Wi-Fi and BLE to avoid cross-modulation.


PCB Layout Rules for Ultra-Low-Jitter Clocks

✅ Do’s:

  • Route clock traces on inner layer, sandwiched between solid ground planes (stripline)
  • Keep trace <5 cm long, impedance-controlled if differential (100Ω diff)
  • Avoid vias under XO or clock IC—they add parasitic inductance/capacitance
  • Place decoupling caps (100 nF + 10 pF) within 3 mm of XO power pins
  • Connect XO case (if metal) directly to local ground island, then to main GND via single point

❌ Don’ts:

  • Run clock traces parallel to switching nodes (SW, BOOT, PHASE)
  • Place XO near USB, Ethernet, or motor drivers
  • Use long stubs or T-junctions on clock nets
  • Share XO ground with digital I/O return currents

🛠️ Bonus: Simulate clock path with SI/PI tool (e.g., Ansys SIwave)—check for resonances near XO frequency.


Real Case: Fixing Wi-Fi Throughput Drop in a Tri-Radio Gateway

Client: Industrial IoT gateway (Wi-Fi 6 + BLE 5.3 + Cat-M1)
Symptom:

  • Wi-Fi throughput dropped from 600 Mbps to 320 Mbps when LTE transmitted
  • No RF desense observed—coexistence filters were adequate

Root cause (via phase noise analyzer):

  • 40 MHz XO showed 3.2 ps RMS jitter during LTE burst
  • Noise peaked at 1.8 MHz—matching LTE PA switching frequency
  • XO powered by shared LDO also used by LTE baseband

Solution:

  1. Added dedicated ultra-low-noise LDO (LT3045) for XO
  2. Moved XO to opposite corner of board, added ground guard ring
  3. Replaced CMOS XO with differential LVDS XO (Rakon RXD-40.000)
  4. Shortened clock trace to 28 mm, routed on Layer 3 (between GND planes)

Result:

  • Jitter reduced to 0.78 ps RMS (12 kHz–20 MHz)
  • Wi-Fi throughput stable at 580+ Mbps during LTE full-power TX
  • Passed 3GPP TS 37.141 intermodulation test
  • BOM impact: + $ 0.85/unit Tested in ChipApex RF & Timing Lab with Keysight E5052B signal source analyzer. — ## Low-Jitter Component Selection Guide | Component Type | Recommended Part | Key Spec | |—————-|——————|——–| | XO (CMOS) | ECS-2520MVLC, TXC 7M | ﹤1 ps RMS, PSRR ﹥40 dB @1 MHz | | XO (LVDS) | Rakon RXD, SiTime Elite | 0.3–0.6 ps RMS, differential output | | LDO | TPS7A4700, LT3045 | PSRR ﹥70 dB @100 kHz, ﹤0.8 µV RMS noise | | MCU/SoC | ESP32-C6, TI CC1352P7 | Integrated low-jitter RF PLL, separate clock domains | ﹥ 💡 Cost note: A 0.90LVDSXO+0.90LVDSXO+ 0.60 LDO often saves $ 5+ in RF rework later.

Common Jitter Mitigation Myths

❌ “A crystal is better than an XO.”
→ Discrete crystals require careful load capacitance matching—and are more sensitive to layout noise. XO integrates optimized driver.

❌ “More decoupling caps = cleaner clock.”
→ Poorly placed caps create LC resonances that amplify noise at specific frequencies.

❌ “Our SoC has built-in jitter cleaning.”
→ Most integrated PLLs lack the performance for sub-1ps applications. External XO is safer.

❌ “Jitter only matters for high-speed SerDes.”
→ Even BLE advertising packets fail if clock drift exceeds ±50 ppm over temperature.


Quick Jitter Readiness Checklist

Before prototyping:

  • XO powered by dedicated ultra-low-noise LDO
  • XO placed far from RF/high-noise blocks
  • Clock trace short, shielded, no vias/stubs
  • Ground under XO is solid and quiet
  • PLL loop bandwidth optimized per datasheet
  • XO datasheet specifies integrated jitter (12k–20M Hz)

🧪 Bonus: Validate with phase noise plot—look for spurs at switching frequencies.


Final Advice from Our FAE Team

“In wireless, you don’t lose bits—you lose time. Control jitter, and you control performance.”
— Mr. Hong, Senior Field Application Engineer, ChipApex


Need Help Achieving Sub-1ps Jitter?

We provide:

  • Low-jitter oscillators (CMOS, LVDS, MEMS) with phase noise plots
  • FAE timing review: Send us your clock tree—we’ll simulate jitter budget
  • Reference designs: Dual-radio IoT, 5G CPE, TSN gateway
  • Lab testing: Phase noise, EVM vs. jitter correlation

Contact Our FAE Team


About the Author

Mr. Hong is a Senior Field Application Engineer at ChipApex with 12+ years in high-speed and wireless system design. He specializes in signal integrity, clocking architecture, and RF coexistence for IoT, automotive, and industrial applications. He holds certifications in high-speed digital design (IEEE) and has supported products through Wi-Fi Alliance, Bluetooth SIG, and 3GPP certification.

The prev: The next:

Related recommendations

Expand more!