⚡ electrical engineering

Proof of Concept: Local TX Real-Time Phase Calibration in MIMO Systems

This paper presents and validates a simple local real-time phase calibration method for digital MIMO arrays, demonstrating that both instantaneous and smoothed approaches effectively achieve sub-picosecond jitter levels on common SDR platforms to enable coherent transmission and advanced beamforming techniques.

Original authors: Carl Collmann, Ahmad Nimr, Gerhard Fettweis

Published 2026-02-19

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Carl Collmann, Ahmad Nimr, Gerhard Fettweis

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The "Choir" Problem

Imagine you are the conductor of a massive choir. Your goal is to get 6 singers (the radio antennas) to sing a single, perfect note together so that the sound travels far and loud in one specific direction. This is called beamforming.

However, there's a problem. Even if you give all 6 singers the exact same sheet music and the same starting cue, their voices will naturally drift apart over time. One singer might get slightly faster, another might get slightly slower, and their pitch might wobble. In the world of radio, this is called phase noise and drift.

If the singers aren't perfectly in sync, their voices cancel each other out instead of amplifying. The signal becomes weak and messy. This paper is about a new, simple way to keep these "singers" perfectly in sync in real-time.

The Old Way vs. The New Way

The Old Way (The "Blind" Approach):
In the past, engineers tried to fix this problem after the signal was received (like a listener trying to fix a bad recording). But for a transmitter (the choir), you can't wait until the sound reaches the audience to fix the singing. You have to fix it while they are singing.

Other methods tried to use a central computer to measure everyone, but that required complex wiring or was too slow for mobile phones.

The New Way (The "Local Conductor"):
The authors propose a clever, local solution. Imagine one of the singers is also a microphone (a reference chain).

The Check-In: Before the choir starts singing the main song, each singer takes a quick turn to hum a short "test note" (a synchronization signal) into the microphone.
The Calculation: The conductor (the controller) listens to these test notes. He hears, "Ah, Singer 3 is slightly sharp, and Singer 5 is slightly flat."
The Correction: Before the real song starts, the conductor whispers a tiny adjustment to each singer: "Singer 3, slow down your breath by a fraction; Singer 5, speed up."
The Result: Now, when they all sing the main song together, they are perfectly aligned.

The paper tests two ways to do this whispering:

Instantaneous: The conductor listens to the very last test note and adjusts immediately.
Smoothed: The conductor listens to the last 10 test notes, averages them out, and then adjusts. This is like taking a "rolling average" to ignore tiny, random mistakes and focus on the real trend.

What They Found (The Results)

The researchers tested this on real hardware (SDRs, which are like programmable radios) and found some fascinating things:

1. The "Warm-Up" Surprise
Even though all the radios were plugged into the same master clock (like a shared metronome), they still drifted apart!

Analogy: Imagine 6 identical watches. Even if they are set to the same time, if you leave them in a cold room and then bring them into a warm room, the heat makes the metal expand, and they start ticking at slightly different speeds.
The Finding: The radios drifted by up to 25 degrees of phase just because they were warming up. This proves you must calibrate them constantly, even if they share a reference.

2. The Precision is Mind-Boggling
The method worked incredibly well. They measured the "jitter" (how much the timing wobbled) and found it was as low as 124 femtoseconds.

Analogy: A femtosecond is to a second what a second is to 31.7 million years.
To put it another way: If you synchronized a clock to this level of precision, and that clock had been running since the Big Bang, it would only be off by a fraction of a second today.

3. Smoothing Wins
The "Smoothed" approach (averaging the last 10 checks) was slightly better than the "Instant" approach.

Analogy: If you are trying to guess the temperature of a room, looking at the thermometer for one split second might show a glitch. If you look at it for 10 seconds and take the average, you get the true temperature. The smoothed method filters out the "noise" to find the real drift.

4. The "Gaussian" Gift
Before calibration, the errors were chaotic and unpredictable. After calibration, the errors became a perfect "Bell Curve" (Gaussian distribution).

Meaning: The system became predictable. The remaining errors were just tiny, random static, which is the best you can hope for in physics.

Why Does This Matter?

This isn't just about making a better radio; it's about the future of 6G and Sensing.

Beamforming: To send data super fast to a specific person without wasting energy, you need to aim the radio beam like a laser. This requires perfect synchronization.
Sensing: Future networks will act like giant radars to detect cars, people, or weather. If the "singers" aren't in sync, the radar image will be blurry.

The Bottom Line

The authors proved that you don't need expensive, complex equipment to keep radio antennas in sync. You just need a simple, local loop where the transmitter checks its own timing constantly and makes tiny adjustments.

They turned a chaotic, drifting choir into a perfectly harmonized orchestra, achieving a level of precision that allows us to build the next generation of wireless communication and sensing systems.

1. Problem Statement

In Multiple-Input Multiple-Output (MIMO) systems, particularly those utilizing digital antenna arrays for beamforming and Joint Communication and Sensing (JCnS), phase coherence across all transmit (TX) Radio Frequency (RF) chains is critical.

The Challenge: While time and frequency synchronization are well-established, maintaining real-time phase coherence is difficult due to phase noise and drift in local oscillators (LOs).
Limitations of Existing Methods:
- Receiver-side compensation: Standard methods (e.g., 3GPP PT-RS) compensate for phase noise at the receiver but do not ensure coherence at the transmitter, which is required for transmit beamforming.
- Over-the-Air (OTA) Reciprocity: Existing OTA methods often lack a shared phase reference or require wired connections between distributed units, making them unsuitable for mobile scenarios or requiring significant overhead.
- Residual Errors: Neglected residual transmit-side calibration errors can substantially degrade MIMO throughput and beamforming gain.

2. Methodology

The authors propose and validate a local, real-time phase calibration method for digital transmit arrays using Software-Defined Radios (SDRs).

System Architecture

Setup: A digital array with $M$ transmit chains and a dedicated reference RF chain located at the transmitter.
Synchronization: A common 10 MHz reference and a trigger signal (e.g., 1-pulse-per-second) align sampling clocks.
Calibration Procedure (TDMA Scheme):
1. Transmission: Each TX chain transmits a synchronization preamble (Phase Tracking Reference Signal - PT-RS) in a Time-Division Multiple Access (TDMA) sequence.
2. Reception: The dedicated reference chain receives the sum of these signals via a cable (isolating channel effects).
3. Estimation: The system estimates the phase of each chain relative to the reference by multiplying the received baseband signal with the conjugate of the transmitted signal.
4. Precoding: The transmission controller applies a phase pre-compensation factor ( $e^{-j\hat{\theta}_m}$ ) to the baseband data signals before transmission to align all chains coherently.

Calibration Approaches Compared

The paper evaluates two strategies for applying the estimated phase:

Instantaneous Calibration: Uses the most recent phase estimate for precoding.
Smoothed Calibration: Uses the average of the last 10 phase estimates to reduce estimation noise.

Modeling

The authors model phase noise for two oscillator types:

Free-Running VCO: Modeled as a Wiener process (random walk), leading to unbounded phase drift.
Phase-Locked Loop (PLL): Modeled as a control loop suppressing long-term drift, where residual jitter is approximated as white Gaussian noise.

3. Key Contributions

Local Real-Time Calibration: A practical, low-complexity method to calibrate TX chains internally without requiring external distributed sensors or complex OTA reciprocity setups.
Quantitative Validation: Introduction of two specific metrics to assess performance:
1. RMS Cycle-to-Cycle Jitter: To measure timing stability.
2. Average Beamforming Power Loss: To measure the impact on array gain.
Hardware Validation: Experimental validation using six USRP X310 chains across different bandwidths and oscillator configurations.
Theoretical Bounds: Demonstration that the proposed method achieves the fundamental performance limits (Cramer-Rao lower bounds) for both VCO and PLL systems.

4. Experimental Results

Hardware Measurements (USRP X310)

Thermal Drift: Significant phase drift (up to $25^\circ$ ) was observed during warm-up, even with a shared 10 MHz reference, highlighting the necessity of continuous calibration.
Jitter Reduction:
- PLL-based chains: Achieved extremely low jitter, reducing RMS cycle-to-cycle jitter from ~193 fs (uncalibrated) to 124 fs (smoothed calibration).
- VCO-based chains: Reduced jitter from ~7.78 ps to 632 fs (smoothed calibration).
- Range: Overall RMS jitter was reduced to the 124 fs – 2.1 ps range depending on the SDR model.
Distribution Analysis:
- Before Calibration: Phase distributions were non-Gaussian (due to drift).
- After Calibration: Residual phase errors became Gaussian distributed (confirmed via Q-Q plots), indicating the calibration successfully "whitened" the phase noise.
Performance Comparison: Smoothed calibration consistently outperformed instantaneous calibration, particularly for chains with higher noise, by averaging out estimation errors.

Simulation Results

Observation Interval ( $t_{obs}$ ):
- For VCOs, periodic calibration is essential. Without it, beamforming gain is lost as the interval increases. With calibration, the system approaches the theoretical VCO limit ( $\sqrt{c_{VCO}t_{obs}}$ ).
- For PLLs, the system is stable over short intervals ( $<100$ ms). Periodic calibration provides marginal gains but is still beneficial for minimizing residual jitter to the reference oscillator limit ( $\sqrt{c_{REF}t_{obs}}$ ).
Beamforming Gain: The simulations confirmed that without calibration, VCO-driven arrays fail to form a coherent beam. With the proposed calibration, a clear main lobe is restored, approaching ideal gain.

5. Significance and Conclusion

Feasibility of SDR Testbeds: The work proves that commonly available SDR platforms (like USRP X310) can achieve the high phase coherence required for advanced MIMO techniques and transmit beamforming, provided a local calibration loop is implemented.
Practical Impact: The method enables precise parameter estimation (e.g., Angle-of-Arrival/Departure) and robust beamforming in practical testbeds, which is crucial for 6G research and Joint Communication and Sensing (JCnS).
Future Work: The authors suggest extending this to joint transmitter-receiver calibration and testing under mobile conditions where Doppler effects are significant.

In summary, the paper demonstrates that a simple, local, time-division multiplexed calibration scheme effectively mitigates transmit-side phase impairments, achieving femtosecond-level jitter stability and enabling coherent transmission in digital MIMO arrays.