A.K.03/student_code/my_firstborn.py

import numpy as np                          # For working with numpy arrays
import matplotlib.pyplot as plt             # For plotting for tests

from scipy.io import wavfile                # For reading .wav files for testing
from scipy.signal import find_peaks         # For cropping the microphone recordings
from scipy.fft import fft, ifft, fftshift   # For channel estimation
from scipy.optimize import least_squares    # For estimating KITT's location

def recording_crop_normalize(recordings, ref_mic):
    # Finding the last peak in the recording of the chosen reference microphone
    ref_sig = recordings[:,ref_mic]
    ref_peaks, _ = find_peaks(ref_sig, height= 0.5*max(abs(ref_sig)))
    ref_peak = ref_peaks[-1]

    # Cropping all recordings to show only the peaks around the reference peak
    start = ref_peak - 3600
    end = ref_peak + 3600
    recordings = recordings[start:end]

    # Normalizing all recordings after they are cropped
    samples, mic = recordings.shape
    recordings_cropped_normalized = np.zeros((samples, mic))
    for i in range(mic):
        recordings_cropped_normalized[:, i] = recordings[:, i]/max(abs(recordings[:, i]))
    recordings = recordings_cropped_normalized
    return recordings

def channel_estimation(recording, reference_recording, epsilon):
    # Finding both the recording and the reference recording in the frequency domain
    padded_length = max(len(recording), len(reference_recording))
    rec_freq = fft(recording, padded_length)
    ref_rec_freq = fft(reference_recording, padded_length)

    # Performing the deconvolution in the frequency domain
    ch_est_freq = (ref_rec_freq*np.conj(rec_freq))/(np.abs(rec_freq)**2+epsilon)

    # Finding the channel estimation in the time domain and centre it
    channel_estimate = abs(ifft(ch_est_freq))
    channel_estimate = fftshift(channel_estimate)
    return channel_estimate

def distance_calc(channel_estimate, sampling_rate):
    # Finding the location of the peak in the channel estimate relative to the reference peak
    center = len(channel_estimate)//2
    peak = np.argmax(abs(channel_estimate))
    sample_range = peak - center

    # Calculating the distance using the Time Difference of Arrival (TDOA) from found peak location
    time_dif = sample_range/sampling_rate
    distance = time_dif * 34300 # cm
    return distance

def location_estimation(mic_locations, ref_mic, distances, start_point = None):
    # Choose a start point which serves as your starting "guessed" location. If no start point is given, choose the middle of the field as the start point.
    if start_point is None:
        start_point = [230,230,0]

    # Using the location of the reference microphone as the refence point
    ref_point = mic_locations[ref_mic]

    # Generating the residuals function that is to be minimized. This residual is the difference between the "guessed" location and the location calculated from the microphone recordings
    def residuals_function(guess):
        guess = np.array([guess[0],guess[1],0])
        residuals = []
        mic, axs = mic_locations.shape
        for i in range(mic):
            if i != ref_mic:
                mic_location = mic_locations[i]
                residual = (np.linalg.norm(guess-mic_location) - np.linalg.norm(guess-ref_point)) + distances[i]
                residuals.append(residual)
        return residuals

    # Using the least squares method to minimize the residuals function
    location = least_squares(residuals_function, start_point, bounds = ([0,0,-1],[460,460,1]))
    return location.x

def localization(recordings, sampling_rate):
    # Choosing a reference microphone. 0 is mic 1; 4 is mic 5
    ref_mic = 4

    # Normalize and crop the recordings
    recordings = recording_crop_normalize(recordings, ref_mic)

    # Finding the channel estimates between each recording and the reference recording
    epsilon = 0.0001
    channel_estimates = []
    recording, mic = recordings.shape
    for i in range(mic):
        if i != ref_mic:
            channel_estimates.append(channel_estimation(recordings[:, i], recordings[:, ref_mic], epsilon))

    # Finding the distances that correspond to the Time Difference of Arrival (TDOA) for each channel estimate
    distances = []
    for i in range(len(channel_estimates)):
        distances.append(distance_calc(channel_estimates[i], sampling_rate))

    # Estimating the location using the least squares method
    mic_locations = np.array([
        [0, 0, 25],  # mic 1 cm
        [0, 460, 25],  # mic 2 cm
        [460, 460, 25],  # mic 3 cm
        [460, 0, 25],  # mic 4 cm
        [0, 230, 55]  # mic 5 cm
    ])
    location_estimate = location_estimation(mic_locations, ref_mic, distances)
    return location_estimate

# Test
if __name__ == "__main__":
    # Coordinates of the recordings
    record_x = [64, 82, 109, 143, 150, 178, 232]
    record_y = [40, 399, 76, 296, 185, 439, 275]

    # Generating the filenames
    filenames = []
    for i in range(len(record_x)):
        real_x = record_x[i]
        real_y = record_y[i]
        filenames.append(f"../files/Student Recordings/record_x{real_x}_y{real_y}.wav")

    for i in range(len(filenames)):
        sampling_rate, recordings = wavfile.read(filenames[i])
        print(f"\nRecording {i+1}: {filenames[i]}")
        location_estimate = localization(recordings, sampling_rate)
        print("Estimated source position:", location_estimate)