基于Gnuradio与LimeSDR的GFSK接收机重采样比特采样位置异常问题求助

Hey there! Let's work through this frustrating 15% packet loss issue you're facing. It's clear the core problem is that your Rational Resampler is occasionally sampling on bit edges instead of centers—here are concrete, actionable fixes tailored to your GNU Radio setup:

1. Ditch the Raw Rational Resampler for Symbol Synchronization Modules #

Your initial thought about Symbol Sync was on the right track, but you don't need to generate a triangle wave manually! For GFSK signals, GNU Radio has built-in modules that handle timing recovery automatically:

• Use the Symbol Sync (Gardner) module: This algorithm tracks symbol timing by analyzing phase differences between consecutive samples, no external reference signal needed. Place it after your GFSK demodulator (but before any resampling) and set the symbol rate to match your transmission rate. It will output exactly one sample per symbol, aligned to the bit center.
• Alternative: Clock Recovery MM module: If you prefer a decision-directed approach, this module uses detected bits to adjust the sampling clock. Pair it with a Binary Slicer after demodulation, and it will lock onto the optimal sampling point for each bit.

2. Add Timing Compensation Before Rational Resampling (If You Want to Keep It) #

If you need to stick with the Rational Resampler, don't use it alone. Add a timing recovery step first to fix the edge-sampling issue:

• Insert the Clock Recovery MM module right before the Rational Resampler. Configure its symbol rate and start with a small loop bandwidth (like 0.01) for stability. This module will dynamically adjust the sampling phase to ensure you're hitting bit centers, so when the Resampler downsamples to 1 sample per bit, it's picking the correct point every time.

3. Optimize Your Baseband Filtering to Reduce Interference #

Bit edge ambiguity often stems from poor filtering leading to inter-symbol interference (ISI). Tweak your filtering stage to clean up the signal before sampling:

• Replace any generic low-pass filter with a Raised Cosine Filter (match this at both transmit and receive ends). Set the roll-off factor to ~0.35 (a standard value for GFSK) to minimize ISI while keeping bandwidth manageable. This will sharpen bit edges and clarify the eye diagram, making it easier for timing recovery modules to lock onto the correct sampling point.
• Use the Eye Diagram Sink in GNU Radio to visualize your signal—if the eye is closed or blurry, your filtering needs adjustment.

4. Validate with Debug Sinks #

To confirm your fixes are working, add these debug modules to your flowgraph:

• QT GUI Time Sink: Compare the signal before and after timing recovery/resampling. You should see clean, aligned bits at the output, no more edge sampling.
• Dual File Sink modules: Save both pre-resampled and post-resampled signals, then cross-check them in MATLAB to verify every bit is correctly sampled once timing modules are in place.

Bonus: Boost MATLAB's Preamble Detection Robustness #

While fixing the resampling issue is the root solution, you can also make your MATLAB preamble detection more forgiving:

• Use cross-correlation with a preamble template that accounts for small timing offsets. Even if there's a tiny sampling shift, a well-designed correlation algorithm can still pick up the preamble.

内容的提问来源于stack exchange，提问作者SomeEngineer100