Channel Impulse Response (CIR): CIR characterizes how a radio channel affects a transmitted signal over time due to multi-path reflections. Understanding CIR is essential for equalizing the signal at the receiver.
RMS Delay Spread & Doppler Shift: RMS delay spread quantifies the multi-path nature of the channel, while Doppler shift accounts for the relative motion between the transmitter and receiver, causing frequency shifts.
Bandwidth: Bandwidth refers to the range of frequencies occupied by a signal. While it influences Channel Capacity (the maximum data rate), they are distinct terms. Telecom evolutions (4G/5G) focus on increasing bandwidth and using higher-order modulation to boost capacity.
Reliability (BER and SER): Bit Error Rate (BER) and Symbol Error Rate (SER) are key metrics for measuring system performance. A 'BER vs. SNR' graph is the standard way to visualize how reliable a system is under different noise conditions.
Propagation Scenarios:
• UMi (Urban Micro): Small cells with ranges of 100-200 meters.
• UMa (Urban Macro): Large area coverage, typically 200m to 2km.
• RMa (Rural Macro): Wide area coverage spanning several kilometers.
Environmental Factors: Pathloss is affected by Line-of-Sight (LOS) or Non-Line-of-Sight (NLOS) conditions, as well as atmospheric parameters like Humidity (default 50%), Temperature, and Rain Rate.
Antenna Arrays:
• Array Types: Common types include ULA (Uniform Linear Array) and URA (Uniform Rectangular Array).
• Spacing: Adjacent elements are typically spaced at 0.5Ī» (half-wavelength) to prevent interference and grating lobes.
Beamforming Parameters:
• AOA/AOD: Angle of Arrival and Angle of Departure of the signal rays.
• HPBW: Half-Power Beamwidth, the angular width where the antenna's power is half of its peak value.
Wiener Filtering & MMSE Estimation
Q. What is the most general frequency-domain expression for the Wiener filter? A. For stationary processes, the general Wiener filter $H(f)$ is expressed using the cross-power spectral density ($S_{dx}$) and the input power spectral density ($S_x$): $$H(f) = \frac{S_{dx}(f)}{S_x(f)}$$ This minimizes the mean-square error between the desired signal $d(n)$ and the estimated signal $\hat{d}(n)$. Q. How are the optimal coefficients computed for an FIR Wiener filter? A. The optimal coefficient vector $h_{opt}$ is found by solving the Wiener-Hopf Equation: $$\mathbf{h}_{opt} = \mathbf{R}^{-1} \mathbf{P}$$ Where $\mathbf{R}$ is the autocorrelation matrix of the input signal and $\mathbf{P}$ is the cross-correlation vector between the input and the desired signal. Q. What is the "Orthogonality Principle" in Wiener filtering? A. The orthogonality principle states that for the optimal filter, the estimation error must be orthogonal to the input data. Mathematically, the expected value of the product of the error and the input signal is zero: $E[e(n) \cdot x(n-k)] = 0$. Q. How does the Wiener filter simplify for uncorrelated additive noise? A. In the model $x(n) = s(n) + v(n)$, if the signal $s(n)$ and noise $v(n)$ are uncorrelated, the filter becomes: $$H(f) = \frac{S_{ss}(f)}{S_{ss}(f) + S_{vv}(f)}$$ In this form, the filter acts as an SNR-dependent gain: it preserves frequencies where the signal power $S_{ss}$ is high and attenuates where the noise power $S_{vv}$ is high. Q. What are the practical limitations of a Non-Causal Wiener filter? A. A non-causal Wiener filter requires knowledge of the "future" values of the input signal to achieve the absolute minimum MSE. While theoretically superior, it cannot be implemented in real-time. For real-time applications, we must constrain the filter to be Causal (using only past and present data).Implementation & Matrix Stability
Q. What is a major challenge in implementing FIR Wiener filters using sample data? A. The primary challenge is the inversion of the sample autocorrelation matrix $\mathbf{R}$. If the input signal is "ill-conditioned" (e.g., highly correlated or limited in variety), the matrix may be nearly singular, making its inverse unstable and highly sensitive to noise. Q. What does the term $\mathbf{P}^T \mathbf{R}^{-1} \mathbf{P}$ represent? A. This term represents the Power Reduction in the estimation error. In the MMSE formula, $J_{min} = \sigma_d^2 - \mathbf{P}^T \mathbf{R}^{-1} \mathbf{P}$. It shows how much of the desired signal's power we can successfully recover/estimate from the observations. Q. When does the optimal Wiener filter become a simple scalar gain? A. This happens when the autocorrelation matrix $\mathbf{R}$ is diagonal. This occurs when the input signal is white (uncorrelated), meaning each sample provides independent information, and the filter doesn't need to consider time-dependencies. Q. Why is the FIR Wiener filter more common than the IIR version? A. FIR filters are always stable and easier to compute via the matrix-based Wiener-Hopf equations. While IIR filters can theoretically provide better performance with fewer coefficients, they are harder to design for causality and can suffer from instability during real-time adaptation.Spectral Estimation & PSD Analysis
Q. Why is the standard Periodogram considered an "inconsistent" estimator? A. In statistics, an estimator is consistent if its variance goes to zero as the number of data points ($N$) increases. For a periodogram, as $N$ increases, the variance remains constant. The estimate simply becomes more "jagged" rather than converging to the true PSD. Q. What is the fundamental trade-off when using a Window function (e.g., Hamming, Hann) in spectral estimation? A. The trade-off is between Resolution and Spectral Leakage.• Windowing reduces sidelobe levels (leakage), which prevents strong signals from masking weak neighbors.
• However, windowing widens the main lobe, which reduces the frequency resolution (ability to distinguish two close peaks). Q. How does the Bartlett Method reduce the variance of the PSD estimate? A. The Bartlett method divides the data into $K$ non-overlapping segments, computes the periodogram for each, and averages them. This reduces the variance by a factor of $K$. Trade-off: Because each segment is shorter, the frequency resolution is reduced by a factor of $K$. Q. What makes Welch's Method the "Industry Standard" for PSD estimation? A. Welch's method improves on Bartlett's by:
1. Allowing segments to overlap (usually 50%), which increases the number of segments averaged and further reduces variance.
2. Applying a window to each segment before the FFT to control spectral leakage. In MATLAB/Python, this is the 'pwelch' function. Q. How does the Blackman-Tukey method differ from the Periodogram? A. The periodogram estimates PSD directly from the data (FFT). The Blackman-Tukey method estimates the Autocorrelation function ($R_{xx}$) first, applies a "lag window" to smooth it, and then takes the FFT. This reduces variance by smoothing the unreliable high-lag components of the correlation. Q. What fundamental assumption is required to estimate PSD from a single signal realization? A. The signal must be Ergodic in the correlation and Wide-Sense Stationary (WSS). WSS ensures the statistical properties don't change over time, and ergodicity allows us to substitute "time averages" for "ensemble averages." Q. Why does a Rectangular Window cause the most spectral leakage? A. A rectangular window (abrupt truncation) has very high sidelobes ($-13$ dB). In the frequency domain, this is a convolution with a Sinc function, which causes the energy of a single frequency to "leak" into almost all other frequency bins. Q. Does Zero-Padding improve the frequency resolution of a PSD estimate? A. No. Zero-padding only interpolates the existing spectrum, making it look smoother (it provides more bins). Actual resolution is determined solely by the length of the data record ($1/N$). No amount of zero-padding can help you distinguish two peaks that are closer than $1/N$.