Here is an eclectic work selection from my oeuvre. I have carried out and managed deliverables as a leader and as an individual contributor to group projects.
Saudi Arabian Interbank Offered Rate (SAIBOR) Forecast
SAIBOR is the mean of interest rates of short-term unsecured loans between national banks in Saudi Arabian Riyals. Published daily by Saudi Arabian Monetary Agency.
- Collected and analyzed more than a 10-year worth of historical data of the daily SAIBOR.
- Investigated its correlation with other international rates as well as some commodity prices during the corresponding period.
- Used auto-regressive models and Long-Short Term Memory (LTMS) for forecasting.
- Performed gird-search across models to optimize and fine-tune predictions
- Incorporated results in estimating the Sharp-Ratio to determine: a) the efficacy of a trading strategy of an AI trader compared to other safer investments and b) market and stock volatility.
5G Massive MIMO
At the cost of having more real estate to accommodate base stations with Massive MIMO, communication networks have never had such a quantum leap in overall performance and spectral efficiency.
Researched extensively Multiple-Input Multiple-Output (MIMO) systems and models. Studied MIMO systems as base stations scale up its antenna arrays, serve more users, and reduce complexity of the RF chain. Which poses some modeling and deployment challenges as well as opportunities for solving some of the longstanding issues in communication networks such as:
- Implementing a cross-layer design to simplify the MAC layer by integrating the pre-coding in the physical layer, resource allocation, and simultaneous channel access.
- Find solutions for the Peak-to-Average Power Ratio and dispense the need for using non-linear power amplifiers in the RF chain.
- Solve the pilot contamination phenomenon where the network broadcast directive interference in the downlink due to the exhaustion of unique pilot sequences.
- Designed and modeled some deployment scenarios and channel estimation schemes that improve the overall performance in both narrow-band and wide-band, aka millimeter-wave band.
- Find solutions for the Peak-to-Average Power Ratio and dispense the need for using non-linear power amplifiers in the RF chain.
- Solve the pilot contamination phenomenon where the network broadcast directive interference in the downlink due to the exhaustion of unique pilot sequences.
- Designed and modeled some deployment scenarios and channel estimation schemes that improve the overall performance in both narrow-band and wide-band, aka millimeter-wave band.
2D Clustering with GMM, k-means, and DBSCAN
Used metrics such as silhouette, AIC, BIC, DIC to compare the performance some unsupervised clustering models.
Algorithmic AI Stock Trading
Hedge funds have long been set out to beat the market and achieve high returns. AI traders are expected to dominate the hedge fund industry; however, instead of the market this time, other AI traders are the ones to beat.
- Collected through API and web-scraping stock prices and market indicators.
- Used historical data to find the optimal investment portfolio for a given number of stocks and risk appetite of trading.
- Built and back-tested multiple trading algorithms using an ensemble of: Bollinger Bands, Admiral Keltner Breakout, Stochastic Oscillator, Moving Average Convergence/Divergence (MACD).
- Compared and contrasted the returns of each trader given a choice of an investment portfolio and risk appetite.
LTE Network Deployment Planning and Link Budget Analysis for Urban Areas
coverage map of a deployment scenario with a Monte Carlo simulator including power control, interference control, inter-cell coordination, aggregation, and backhaul capacity constraints traffic simulator using Atoll
- Planned the deployment of a 4G LTE network in an urban area according to given specifications: cell average and edge throughputs; SNIR, QoS and sensitivity requirements; radio QoS profiles; interference power levels.
- Simulated the coverage mapping for different base station locations, bands and operating modes.
- Calculated for each simulated deployment scenario LTE Link Budget and MAPL.
- Simulated traffic and optimized the network by anticipating load and traffic patterns.
Deep-fake Images for an Imaginative Family Member with Generative Adversarial Network (GAN)
The vast trove of images available online for the British royal family makes them a perfect candidate for deep-fake GANs
- collected via web-scraping images for members of a famous family.
- run the collection through an automated pre-processing and re-scaling prior to feeding them to the designed generative adversarial network for training.
- run the collection through an automated pre-processing and re-scaling prior to feeding them to the designed generative adversarial network for training.
- Trained a GAN of the processed collection of images.
- Tweaked and tampered with the discriminator model to generate a more specific instance of an deep-fake image, e.g. if princes William and Harry had a sister princess, what would she look like? what would the output look like if we encouraged the generator and discriminator to sample the late princess Diana and tolerate more resemblance to her?
Semi-Blind Channel Estimation for 5G Massive MIMO Networks
Fast-growing wireless communication networks and services bring greater demands for high spectral efficiency.
- Designed a channel estimation algorithm based on the Expectation Maximization algorithm.
- Designed the algorithm to alternate and optimize between training and uplink transmission phases, and time and frequency domains.
- Extracted extra information from data to improve upon the channel estimate obtained from channel training only.
- Modeled and simulated multiple deployment scenarios to find gains in performance.
3D Models for Wirelss Communication Channels
- Researched and investigated the effect of incorporating a realistic antenna-pattern in channel estimation and precoding in contrast to a presumed isotropic antenna element or communication in the azimuth plane.
- Used extra resolution in azimuth and elevation plane to mitigate the front-back and elevation ambiguity problems.
Spread Spectrum Communications
Once proclaimed “The Most Beautiful Woman in Films,” Hollywood actress Hedy Lamarr is also a self-taught inventor whose brainchild the Frequency-Hopping Spread Spectrum Technology would one day form the basis for today’s WiFi, GPS, and Bluetooth communication systems.
Modeled and simulated both the physical and multiple access layers various spread spectrum techniques and their combinations as means to reverse-engineer their jamming-resistant properties and utilize then for other applications. Methods investigated include:
- Frequency-Hopping spread spectrum (FHSS)
- Direct-Sequence Spread Spectrum (DSSS)
- Time-Hopping Spread Spectrum (THSS)
- Chirp Spread Spectrum (CSS)
Dimensionality Reduction with PCA, t-SNE and Auto-encoders
- Studied, built, compared, and contrasted PCA, t-SNE, and Auto-encoders. Reconstructed data and measured the loss of information resulted from the dimensional reduction.
- Used 3D visualization to check for the efficiency of each method on the input dataset.
Recommendation Systems
- Constructed a sparse user-content matrix from a dataset.
- Built recommendation systems using Probabilistic Matrix Factorization (PMF) and Context-Aware, Matrix Factorization.
- Compared result with the output of Surprise library and k-nn.
Pilot Contamination Massive MIMO Networks
Pilot Contamination occurs when a base station reuses a pilot sequence assigned already to another one nearby. The determent to the network performance is twofold: the interfering signal acts as colored noise that reduces the channel estimation accuracy, and the base station unintentionally estimates a superposition of the channel from the desired terminal and the interferer.
Investigated the Pilot Contamination phenomenon, an artifact created when the burden of transmitting pilot signal is shifted to the users and , subsequently, make the network prone to directive interference due to the limited number of pilot sequences. Solutions typically fall under the following categories:
- Cooperative-cell Networks.
- Protocol design.
- Pre-coding and post-processing to signals
- 3D channel modelling and angle-of-arrival estimation.
Mobile Ad-hoc Networks Optimization
- Studied extensively various networks including: IoT, sensor, green, intelligent transportation, and intermittently Networks.
- Implemented and designed different routing and flow and congestion control schemes in a cross layer design to ensure fair queuing and serve according to differently Quality-of-Service profiles.
- Expanded some routing protocol designs to accommodate changing network typologies and avoid deadlocks.
- Pre-coding and post-processing to signals
- 3D channel modelling and angle-of-arrival estimation.
Image Stitching with Radial and Structure Deformation
- Employed gradients and convolution to achieve a seamless and coherent image stitching.
- Convolved images with a gaussian kernel to find feature in each image and match them.
- Used gradient to find the optimal stitching seam.
- Construct polygons whose vertices are the resultant feature for the convolution, deform and feather them accordingly to achieve a seamless stitch.
OCR and Image Classification with Deep Learning
OCR and Image Classification with Deep Learning
- Designed, built, and trained two neural networks: a deep neural network and a convolutional neural network (CNN).
- Models built with TensorFlow.
Cross-Layer Design in Moble Ad-Hoc Networks
Designed, simulated, and tested the performance of multiple proposed cross-layer schemes for different deployment scenarios and applications, including: sensor networks, wireless mesh networks, and IoT.
