Wireless Tech + AI: Key Interview Questions You Should Know

1. Explain OFDM system briefly.

• OFDM is a digital modulation technique used in many modern wireless communication systems, including Wi-Fi (IEEE 802.11ax/be), LTE, and 5G. It efficiently transmits a high data stream over multiple closely spaced orthogonal subcarriers (low data stream).
• Subcarriers are spaced such that they don't interfere with each other, despite overlapping spectrally. This maximizes spectral efficiency.
• OFDM is resilient to multiparty fading and delay spread, due to its use of cyclic prefix (CP).

2. How is OFDM resilient to multipath fading and delay spread?

In wireless channels, signals often reflect off objects like walls, furniture, or people, creating multiple delayed copies of the transmitted signal. This causes ISI (Delayed signals from previous symbols interfere with the current symbol) and frequency selective fading (Some frequencies experience more fading than others due to constructive/destructive interference).

OFDM Solution: Use Cyclic Prefix (CP)

• If CP duration ≥ maximum delay spread of the channel, then all multipath components fall within the CP and don't interfere with the useful parts of the OFM symbol.

3. In an OFDM system, how can we find the start of the OFDM symbol from the received data?

• CP based correlation
• CP is a copy of the last part of the OFDM symbol prepended to the beginning. The receiver correlates the received signal with a delayed version of itself to detect the repeating pattern.
• ML based approach
• Train a Neural Network (NN) model to predict symbol timing from raw samples (used in dynamic channels). 
• Advantage: Adapts to non-ideal channel conditions.

4. Explain a Physical (PHY) layer challenge in 6 GHz WiFi and how AI can address it.

The 6 GHz band introduces WiFi 6E and WiFi 7, enabling high throughput and low latency communication. However, a key challenge at the PHY layer is the efficient beam alignment for high frequencies communication.

• Why Beam alignment:
• High frequencies suffer from high path loss and blockage in an indoor environment.
• Directional communication is very important.
• Overhead in beam training: Traditional exhaustive beam sweeping (testing all possible beam pairs) introduces significant latency and overhead, degrading performance in modern WiFi systems. 
• How AI can help:
• VAE's for beam alignment
• VAE's can help reduce beam search complexity to reduce beam alignment procedure.
• VAE's can be trained on multiple essential features, such as AoA/AoD tensors, permittivity tensors, and Access Point (AP) and receiver location tensors.
• VAE can map these multiple tensors to Channel State Information (CSI) to correlate with optimum beam directions.
• VAE's for fast beam prediction
• Instead of exhaustive search, we can use VAE's decoder to predict the beam pairs, reducing the search space.
• VAE's for dynamic scenarios
• VAE's generalize better than other AI models when faced with dynamic environment setups.
• Unsupervised learning allows adaptation without labeled data.

OFDM resource allocation using LSTM:

• Challenge: In 6 Ghz WiFi, dynamic resource allocation in OFDM systems is a challenge.
• Solution: AI enabled system can optimize subcarrier and power allocation using LSTMs.
• LSTM learns temporal patterns in traffic demands and channel conditions.
• It can predict future resource needs and pre-allocate subcarriers, reducing scheduled latency.
• Example: An LSTM trained on past CSI and QoS metrics can predict optimal resource blocks for each user.

5. How would you optimize 802.11 be multi -AP coordination using ML?

• We can use Federated Learning for distributed AP's.
• I would also like to propose a "Application of Simulated Annealing Algorithm (SAA) to Optimize 3D robotic arm for WiFi AP".

6. Which ML model would you use for dynamic channel access and why?

I will explore 2 options: Recurrent NN (RNN) and LSTM.

• RNN struggle with long sequence, but LSTM memory cells retains past information effectively.
• RNN's become unstable with long-term tracking, but LSTMs remain robust for continuous shelf movement analysis.

7. Explain the IFFT technique used in OFDM.

OFDM systems use IFFT to convert frequency-domain symbols to time-domain symbols after modulating all the subcarriers.

Suppose you have a set of frequency-domain complex symbols:

$$X[0],X[1],X[2],\dots ,X[N-1]$$

These represent data mapped to N subcarriers (after modulation like QPSK, QAM, etc.).

The IFFT generates a time-domain sequence $x[n]$ of length $N$ using the formula:

$$x[n]=\frac{1}{N}\sum _{k=0}^{N-1}X[k]\cdot e^{j2\pi kn\mathrm{/}N},n=0,1,2,\mathrm{.}\mathrm{.}\mathrm{.},N-1$$Example: IFFT with 4 QPSK Symbols

Let’s assume the input frequency-domain vector $X[k]$ contains the following QPSK-modulated complex symbols:

$$X=[1+j,-1+j,-1-j,1-j]$$

These are your modulated subcarriers $X[0],X[1],X[2],X[3]$.

For $N=4$, the IFFT formula is:

$$x[n]=\frac{1}{4}\sum _{k=0}^{3}X[k]\cdot e^{j2\pi kn\mathrm{/}4},n=0,1,2,3$$

We’ll compute each time-domain sample $x[0],x[1],x[2],x[3]:$

1. For $n=0$:

$$x[0]=\frac{1}{4}\sum _{k=0}^{3}X[k]\cdot e^{j0}=\frac{1}{4}(X[0]+X[1]+X[2]+X[3])$$$$x[0]=\frac{1}{4}((1+j)+(-1+j)+(-1-j)+(1-j))=\frac{1}{4}(0+0j)=0$$

2. For $n=1$:

$$x[1]=\frac{1}{4}\sum _{k=0}^{3}X[k]\cdot e^{j2\pi k\mathrm{/}4}=\frac{1}{4}(X[0]\cdot 1+X[1]\cdot e^{j\pi \mathrm{/}2}+X[2]\cdot e^{j\pi }+X[3]\cdot e^{j3\pi \mathrm{/}2})$$

Simplify exponentials:

• $e^{j\pi \mathrm{/}2}=j$,

• $e^{j\pi }=-1$,

• $e^{j3\pi \mathrm{/}2}=-j$.

Now substitute $X[k]$:

$$x[1]=\frac{1}{4}((1+j)(1)+(-1+j)(j)+(-1-j)(-1)+(1-j)(-j))$$$$=\frac{1}{4}((1+j)+(-j-1)+(1+j)+(-j-1))=\frac{1}{4}(2+0j)=0.5$$

3. For $n=2$:

$$x[2]=\frac{1}{4}\sum _{k=0}^{3}X[k]\cdot e^{j2\pi k\cdot 2\mathrm{/}4}=\frac{1}{4}(X[0]+X[1]\cdot e^{j\pi }+X[2]\cdot e^{j2\pi }+X[3]\cdot e^{j3\pi })$$

Simplify exponentials:

• $e^{j\pi }=-1$,

• $e^{j2\pi }=1$,

• $e^{j3\pi }=-1$.

Substitute $X[k]$:

$$x[2]=\frac{1}{4}((1+j)(1)+(-1+j)(-1)+(-1-j)(1)+(1-j)(-1))$$$$=\frac{1}{4}((1+j)+(1-j)+(-1-j)+(-1+j))=\frac{1}{4}(0+0j)=0$$

4. For $n=3$:

$$x[3]=\frac{1}{4}\sum _{k=0}^{3}X[k]\cdot e^{j2\pi k\cdot 3\mathrm{/}4}=\frac{1}{4}(X[0]\cdot 1+X[1]\cdot e^{j3\pi \mathrm{/}2}+X[2]\cdot e^{j3\pi }+X[3]\cdot e^{j9\pi \mathrm{/}2})$$

Simplify exponentials:

• $e^{j3\pi \mathrm{/}2}=-j$,

• $e^{j3\pi }=-1$,

• $e^{j9\pi \mathrm{/}2}=e^{j(4\pi +\pi \mathrm{/}2)}=j$.

Substitute $X[k]$:

$$x[3]=\frac{1}{4}((1+j)(1)+(-1+j)(-j)+(-1-j)(-1)+(1-j)(j))$$$$=\frac{1}{4}((1+j)+(j+1)+(1+j)+(j+1))=\frac{1}{4}(4+4j)=1+j$$

Combining all $x[n]$:

$$x[n]=[x[0],x[1],x[2],x[3]]=[0,0.5,0,1+j]$$

8. How does IFFT ensure orthogonality in OFDM?

Subcarrier spacing Δf=1/Tu guarantees zero inter-carrier interference (ICI).

Comment below if I repeat any interview question. Thank you!