Possible GNU Radio implementation of 2nd order polynomial baseline flattening

In GNU Radio, “2nd order baseline flattening” isn’t a single built-in block—you implement it by estimating a quadratic trend and subtracting it from your signal.

Here are the practical ways to do it in a flowgraph. Note this has not been tested, so I provide no guarentee that it will work!

🔧 Method 1: Python Block (most flexible — recommended)

You fit and remove a 2nd-order polynomial in real time.

Steps

1. Add a Stream to Vector block
   • Set vector length (e.g. 1024 or 4096 samples)
2. Add a Python Block to:
   • Fit a quadratic
   • Subtract it

Example Python code

import numpy as np

from gnuradio import gr

class blk(gr.sync_block):

    def __init__(self, vlen=1024):

        gr.sync_block.__init__(

            self,

            name="2nd_order_baseline_removal",

            in_sig=[(np.float32, vlen)],

            out_sig=[(np.float32, vlen)]

        )

        self.vlen = vlen

    def work(self, input_items, output_items):

        x = np.arange(self.vlen)

        for i in range(len(input_items[0])):

            y = input_items[0][i]

            # Fit 2nd order polynomial

            coeffs = np.polyfit(x, y, 2)

            baseline = np.polyval(coeffs, x)

            # Subtract baseline

            output_items[0][i][:] = y - baseline

        return len(output_items[0])

3. Follow with:
   • Vector to Stream
   • FFT / display blocks

🔧 Method 2: Approximation using filters (simpler, less precise)

If you don’t want Python:

• Use a high-pass filter to remove slow trends
• Or chain:
  • DC Blocker (removes 0th order)
  • High-pass (removes low-frequency curvature)

⚠️ This is not true 2nd-order fitting, but often “good enough” for spectra.

🔧 Method 3: Offline (for best accuracy)

If you’re already saving data (which you are with your interferometry setup):

1. Export to file
2. Use Python/NumPy:

y_corrected = y - np.polyval(np.polyfit(x, y, 2), x)

💡 Practical tips (radio astronomy context)

• Use only baseline regions (exclude strong signals like the 1420 MHz hydrogen line)
• Larger vector sizes = smoother fit
• Don’t overfit (3rd+ order can distort real signals)

🧠 Rule of thumb

• Slight tilt → 1st order
• Gentle curve → 2nd order ✅
• Wavy baseline → filtering or higher order (careful!)