Christopher Kitras
PhD Candidate
Department of Electrical and Computer Engineering
Brigham Young University
Research
I am a PhD candidate in the Department of Electrical and Computer Engineering at Brigham Young University (BYU), advised by Philip Lundrigan. As a network researcher in the NET Lab, I make systems more intelligent, adaptable, and resilient. Before starting my PhD, I completed my B.Sc. in Computer Engineering with a minor in Computer Science at BYU in 2021.
My research interests include sensor networks, IoT resilience, and systems engineering. I am particularly interested in how these fields can be leveraged to enhance the capability of simple systems without adding extra hardware to the architecture.
Current work
-
Now
Connectivity for unassociated devices
In progress
-
2024–26
Latency-based Wi-Fi subprotocol
Paper under review
Publications
-
If You Build It, Will They Connect? 6 GHz WiFi in a Stadium
IEEE International Conference on Computer Communications and Networks (ICCCN), 2026
The opening of the 6 GHz spectrum promises substantial capacity gains for dense WiFi deployments, yet real-world performance depends critically on configuration choices rather than spectrum availability alone. This paper presents a measurement study of a large-scale outdoor WiFi 6E deployment at Brigham Young University’s LaVell Edwards Stadium, combining two years of operational data with targeted measurements during football games. Despite the RF advantages of 6 GHz, capacity-based configuration preferences result in only modest performance improvements compared to 5 GHz. While 6 GHz achieves 3.4× higher median throughput (28 Mbps vs 8 Mbps) and 90% lower channel utilization (3.1% vs 31.4%), both bands fall well below expected performance levels. More critically, client adoption remains the dominant limiting factor: automated Passpoint steering redirects ∼50% of stadium clients, an estimated 20% of whom are 6 GHz-capable, exclusively to 5 GHz, while WiFi chipset-level association decisions prevent manual band selection. Historical data reveals that migration to OWE transition mode increased 6 GHz adoption from near-zero to 21% of non-steered clients.
Illustrative.Where phones end up. About half the crowd is signed in to Verizon's 5 GHz network by Passpoint before any band choice happens, including phones that could use 6 GHz (rings). Of the rest, 21% reached 6 GHz once OWE transition mode let them move automatically in 2025. Switch to 2024, when fans had to choose the 6 GHz network by hand, and almost nobody did. The bar shares follow the paper's estimates; the 2024 value and the motion are illustrative. -
Long-Term Validation of a Consumer- vs. Industrial-Grade Indoor Radon and Air Quality Monitor at a School in Utah
Sensors, vol. 26, no. 16, article 5230, 2026
In recent years, the demand for indoor radon and air quality monitors has increased while more affordable consumer-grade monitors have become available. The aim of our study was to assess the validity of the less expensive, non-calibrated, consumer-grade Lüft (SunRADON LLC, Melbourne, FL, USA) compared to the more expensive, annually and manufacturer calibrated, industrial-grade Model 1028 XP (SunRADON LLC, Melbourne, FL, USA). We collected data about 24 h moving averages of barometric pressure, carbon dioxide (CO₂), radon, relative humidity, temperature, and volatile organic compounds (VOC) for approximately 29 weeks in 2024 in each of six rooms in a school in Utah. The mean difference in the co-located Lüfts’ and Model 1028 XPs’ 24 h moving average radon was −0.05 pCi/L of air (95% confidence interval: −0.24, 0.14, p = 0.50) and the concordance correlation coefficient (CCC) was 0.92. The Lüfts’ and Model 1028 XPs’ 24 h moving averages were similar for barometric pressure (CCC: 0.87), but the Lüfts’ 24 h moving averages were lower than the Model 1028 XPs’ for CO₂, relative humidity, and VOC and higher for temperature. The Lüft provided valid measurements of 24 h moving averages for barometric pressure and radon, but not for CO₂, relative humidity, temperature, and VOC.
Illustrative.What a 24-hour moving average keeps. The thin line is hourly radon, the kind of record the 1028 XP keeps. The window slides along it and traces the 24-hour average, which is all the Lüft reports and what the 1028 XP data was converted to for the comparison. Hide the hourly line to see what is lost: the daily dips while the HVAC runs. The hourly series is synthetic; the weekday and weekend pattern follows the one the study observed. -
Sustainable Radon Mitigation through Optimized HVAC Scheduling
ACM/IEEE International Conference on Connected Health: Applications, Systems and Engineering Technologies (CHASE), New York, 2025
Radon is a naturally occurring radioactive gas and is the second leading cause of lung cancer after smoking tobacco. Its invisible and odorless nature often leads to it going undetected, posing a significant health threat in susceptible areas, especially schools. Although traditional mitigation strategies are reasonable for smaller private properties, their cost can become prohibitively expensive and difficult for older and larger buildings. We propose, build, and test a cost-effective, sustainable mitigation strategy that uses the existing HVAC infrastructure of susceptible buildings to remove radon before occupancy hours. We compare this to a naive approach and find that we can save time on HVAC operation and keep radon levels to an acceptable low.
Illustrative.A week of HVAC schedules. A clock sweeps Monday to Friday, filling in when the HVAC runs under each schedule and adding up the hours; the optimized schedule starts at the time in Table 1 and stops at 16:00, so it runs the fewest hours. Runtime is the paper's energy proxy: it reports hours saved, not kWh, and the totals here are computed from the real start and end times. Use Next season and Next zone to step through Table 1. The radon curve shapes are synthetic, drawn so the optimized curve reaches 4 pCi/L at 07:00 as the algorithm intends. -
Hidden in Plain Sight: Communicating using Interference
IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN), 2025
Radio astronomy observatories examine the universe by capturing faint RF signals from space. The equipment to capture these signals is extremely sensitive, and nearby transmitters, such as LEO satellites, cause destructive interference. We propose the Spectra watermarking protocol to add an identifying fingerprint to these interfering transmitters. Spectra changes the timing of key transmissions to encode identifying information that can be decoded by examining transmission timestamps and without having to demodulate the actual transmission. We implement multiple forward error correction codes for Spectra to account for dynamic LEO satellite channels. We evaluate Spectra and show that it can operate over highly variable channels with low BER while preserving the network performance of the interfering device.
Illustrative.Who ruined the observation? A satellite's sidelobe drowns the telescope's faint source, and the observatory names the culprit from transmission timestamps alone, without demodulating anything. Each watermark carries a 16-bit one-time code from a seed the satellite and observatory share, valid for one minute. Press Show a replay attack and Eve records Sat B's watermark, then replays it three minutes later to frame Sat B; the code has expired, so the observatory rejects it. The code length, one-minute slice and timing-only decode are from the paper; satellite paths, burst timing and code values are made up. -
Utilizing Low-Cost Sensors to Monitor Indoor Air Quality in Mongolian Gers
Sensors, vol. 23, no. 18, article 7721, 2023
Air quality has important climate and health effects. There is a need, therefore, to monitor air quality both indoors and outdoors. Methods of measuring air quality should be cost-effective if they are to be used widely, and one such method is low-cost sensors (LCS). This study reports on the use of LCSs in Ulaanbataar, Mongolia to measure PM2.5 concentrations inside yurts or “gers”. Some of these gers were part of a non-government agency (NGO) initiative to improve insulating properties of these housing structures. The goal of the NGO was to decrease particulate emissions inside the gers; a secondary result was to lower the use of coal and other biomass material. LCSs were installed in gers heated primarily by coal, and interior air quality was measured. Gers that were modified by increasing their insulating capacities showed a 17.5% reduction in PM2.5 concentrations, but this is still higher than recommended by health organizations. Gers that were insulated and used a combination of both coal and electricity showed a 19.1% reduction in PM2.5 concentrations. Insulated gers that used electricity for both heating and cooking showed a 48% reduction in PM2.5 but still had higher concentrations of PM2.5 that were 6.4 times higher than recommended by the World Health Organization (WHO). Nighttime and daytime trends followed similar patterns and trends in PM2.5 concentrations with slight variations. It was found that at nighttime the outside PM2.5 concentrations were generally higher than the inside concentrations of the gers in this study, meaning that PM2.5 would flow into the ger whenever the doors were opened, causing spikes in PM2.5 concentrations.
Illustrative.A day and a night inside a ger. Coal smoke from cooking and heating fills the ger, and the sensor on the center beam reads it. At night the air outside is dirtier than inside, so every time the door opens a puff of outdoor PM2.5 floods in. Press the ger-type button to compare traditional, insulated, coal plus electric, and electric-only gers. Each curve's daytime and nighttime averages equal the paper's (Figs. 5 to 7); the hour-by-hour shape, door times and particle motion are made up. -
Location Verification of Crowd-Sourced Sensors
IEEE International Conference on Computer Communications and Networks (ICCCN), Honolulu, 2023
Community-driven sensor networks have been instrumental in providing easy access to affordable, large-scale measurement recording, facilitated by the accessibility of inexpensive sensor hardware. The simplicity of this hardware makes it challenging to retrieve trustworthy location data without added hardware such as GPS. We introduce the LaMDA framework, a software-based solution run solely in a web browser to determine the location of a device with the aid of a registering device. We monitor the device for any location changes by analyzing its traceroute data from a central server. Our solution allows minimal firmware changes to be made to fleets of devices without recall or changes to hardware.
Illustrative.From registration to a moved sensor. The server first checks that the sensor's latency matches the owner's phone to within 15 ms. It then traceroutes the sensor's home router every 10 minutes for a week: links it sees often grow thicker and load-balanced detours branch off, leaving a weighted path graph that is the sensor's signature. After a gap in uploads, new traces are scored against it, 99.7% if the sensor stayed or 11.3% after a move between cities (Table I(b)), against the 90% threshold. Use the button to keep the sensor in place or move it; the hop graph, latencies and timing are illustrative.
Posters
-
Air-quality readings come from a limited number of stations and privately bought sensors, which cluster where people can afford them. TCAM pairs years of images from Utah's public traffic cameras with hourly PM2.5 readings to estimate air quality from the picture alone. We collected over 218 million images from 1,395 cameras, and in a first test on one camera the model picked the right air-quality category 98.6% of the time.
Illustrative.The TCAM data dashboard. As on the poster's dashboard, a 5 km radius at Provo selects the traffic cameras, PurpleAir sensors, DAQ stations and weather stations inside it, and one camera's frame goes to the YOLO-cls model, which picked the right air-quality category 98.6% of the time on one camera. Toggle the source layers, move the radius to another town, or click the map. The map, station positions and Provo query follow the poster's screenshot; camera and PurpleAir positions are a synthetic sample of the 1,395 cameras and 778 sensors, so the in-radius counts count that sample, and the camera frame and its category are synthetic. -
Radon Mitigation through Optimized HVAC Scheduling
Air Quality: Science for Solutions (S4S), 2024 Best Poster Award
Radon collects in school buildings overnight, and conventional mitigation is too costly for large, older buildings. Using radon monitors and the school's existing HVAC system, we scheduled ventilation so radon falls below the 4 pCi/L limit before staff arrive, while running the system less than a fixed schedule does. The full study is the CHASE 2025 paper.
Illustrative.Same low radon, fewer HVAC hours. Bars are the average radon while the school is occupied, per weekday and schedule, from the poster's Health Impact chart; all stay well below the EPA and WHO limits. The optimized start times are the poster's schedule table, and the hours use the paper's 06:00–17:00, 05:00–18:00 and 16:00 end times. Use Next day to compare one weekday at a time. The floor plan is a simplified trace of the school, not to scale. -
SSS: Building a Seven Segment Sign
PyCon US 2023, poster and demo First Place, BYU IoT Pi Competition
A sign made of 1,152 seven-segment digits in a 48 by 24 grid, driven by one Raspberry Pi and written entirely in Python. A scheduler loops demos for the lab until someone takes over from a phone or a game controller, and it plays games, videos and even Doom. It won first place at a BYU Raspberry Pi competition and was shown as a poster and live demo at PyCon US 2023.
Illustrative.One digit, two pixels. The grid is one panel of the sign: 16 × 6 digits, which the library treats as 16 × 12 pixels. As the shape is drawn, the zoomed digit on the left shows which half the current pixel is and the four segments it lights. Switch shapes to draw a wave. The shapes and drawing speed are made up for the figure. -
Location Monitoring Framework for Citizen Science Sensors
Air Quality: Science for Solutions (S4S), 2023 Best Poster Award
Community air-quality sensors have no GPS, so their owners can register them anywhere on a public map. This framework verifies a sensor's location at registration by borrowing the owner's phone location and comparing the phone's and the sensor's latency to a server, then watches the sensor's network path afterward for signs that it was moved. The full system is described in the ICCCN 2023 paper.
Illustrative.The field test, node by node. A sensor is trusted at one node, goes quiet, then reports traces from another node, and its score from the poster's tables lands against the 90% line. Rural and inter-city moves score 23.9% or less; the closest intra-city pairs reach 66.6%, still caught but by a much narrower margin, while each node scores 94.9% to 99.9% against itself. Press Next region to jump between rural, inter-city and intra-city. The scores are real; the maps and node positions are drawn roughly after the poster and are illustrative.
Projects
-
Music Generation with Direct Policy Optimization
CS 673: Computational Creativity, Brigham Young University
Music is a fundamental creative outlet that allows humans to express themselves meaningfully. However, not every human who has the desire to express themselves through music has the skill set to do so effectively. MusicGen is a music generation model that allows anyone to input a text or audio prompt and receive a generated piece of music as output. In this paper, we present a system that allows a potential music composer to utilize MusicGen more intentionally. We enable the user with even more control over the generation of their musical piece by training the MusicGen model against a reward model optimized for certain aspects such as danceability, valence, etc. We accomplish this by using a Spotify Research dataset to train MusicGen with Direct Preference Optimization. We evaluate the success of our work through a set of tests that will determine the variety of the music generated, the compliance of creating music according to the specified parameter configuration given, and the accuracy of our customized models.
Illustrative.One preference pair at a time. Both songs run through MusicGen's decoder. Only the last 12 of its 24 layers change; the first 12 stay frozen so the model keeps what it knows about music. Training raises how likely the model finds the chosen song, compared with the frozen original, and lowers it for the rejected one. The gap is the margin. Swap the pair to train the negative model for the same metric. The songs, values, and steady growth of the margin are invented for clarity; in our real runs most margins barely moved. -
A Replication Study of DSAV Measurement
CS 660: Advanced Computer Networks, Brigham Young University
Destination-side source address validation (DSAV) plays an important role in the modern internet. Despite this, many autonomous systems have a history of this service being disabled, as was indicated in a 2020 measurement study on DSAV. This paper acts as a replication study for that work and seeks to measure any change in DSAV utilization. While little has changed overall in global DSAV utilization between the two studies, several outliers are discovered and discussed.
Illustrative.The measurement. The prober sends one query from each of the five kinds of forged source. With DSAV off, the believable ones reach the resolver, which then looks the name up at our server, and we log the failure. Turn DSAV on and the gateway drops them all. The 2020 and 2024 columns are measured: among targets that leaked at all, the share reached by at least one query of that kind. The addresses and the animation are examples; in the real data, private, destination-as-source, and loopback queries almost never produced a lookup in 2024. -
Unintentional Darknet: An Analysis of Scanning Behavior through the Lens of Abandoned Domains
CS 660: Advanced Computer Networks, Brigham Young University
An abandoned BYU research project's website still draws hundreds of thousands of DNS lookups years later. Using three weeks of the university's DNS logs, we separated hosts that eventually find the project's real address from hosts that keep asking for addresses that lead nowhere. Hosts that never found the real site appeared on a public abuse database far more often (42.9% against 18.5%), which suggests a dead domain can double as a free scanner detector.
Illustrative.Two kinds of visitors. Hosts in the top group fail once, ask for the real site, get an answer, and go quiet because it's cached. Hosts in the bottom group keep cycling through dead names (counters show queries sent). Switch the log to follow either kind. The AbuseIPDB shares are from the report; host counts, timing, and query order are illustrative. -
Chess Sight: A Computer Vision System for Stockfish
ECEN 631: Robotic Vision, Brigham Young University
A webcam watches a real chessboard and lets you play the Stockfish engine on it. The app lines its view up with a reference photo of the board, finds the blue and red pieces by color, and works out each move from which squares emptied and filled. Stockfish checks the move and answers, and its replies are drawn as arrows on the live video.
Illustrative.From camera frame to occupied squares. Feature matches tie the camera frame to the reference photo, RANSAC throws out the bad pairs, and the fitted homography warps the frame so the 64 squares line up. Only blue and red pixels survive the color masks, and each square is summed and thresholded, so the two small specks don't count as pieces. Move the camera to see the same steps from another angle. The camera poses, feature points, and specks are made up; the order of the steps follows the code. -
Interactive Underwater Acoustic Simulator with Ray Tracing
ECEN 528: High Performance Parallel Computing, Brigham Young University
BELLHOP is a standard model of how sound bends underwater as temperature and salinity change with depth, but using it meant editing configuration files and rerunning it for every change. We built an OpenGL program that runs the model inside the render loop and draws its rays live, with on-screen controls for the sound-speed profile. It draws hundreds to thousands of rays in real time, and on a GPU a full frame took about 5 ms instead of 17.
Illustrative.Rays bending through a sound-speed profile. Each ray is marched in small steps and turned at each one by the local sound-speed gradient (Snell's law), the same idea BELLHOP's ray mode uses. The left panel is the profile, as in our simulator's second viewport. Switch profiles to see trapped rays in a deep sound channel, straight rays in uniform water, and rays curving upward when speed rises with depth. The profiles, depths, and ray count are textbook examples, not the environment from the report. -
WiLiE: Wireless LiDAR Explorer, a Cost-Effective Solution to Routine Network Metric Collection
ECEN 526: Wireless Networking, Brigham Young University, 2022
Commercial Wi-Fi survey tools are expensive and depend on a person walking a floor plan and marking where they stand. WiLiE is a remote-controlled robot built from Raspberry Pis, a microcontroller and a low-cost LiDAR that maps the room as it drives and tags every signal-strength reading with where it was taken. In a lab test it placed readings to within about 50 cm and produced heat maps that showed small pockets of weaker signal.
Illustrative.A survey drive. Only what the LiDAR has hit is drawn, so the map builds as the robot moves. Signal-strength samples drop along the path and are blended into a heat map at the end. Switch antennas to see the onboard antenna read the same field weaker. The floor plan, path, and signal field are made up; the two dB ranges on the legend are the ones measured in the project. -
EpicEth: An HDL Circuit to Control a 3D Printer over Ethernet
ECEN 620: Advanced Digital Systems, Brigham Young University
Two FPGA boards drive a 3D printer over Ethernet. Switches on the sending board compose a short move command; the receiving board runs a small TCP server that translates each message into G-code and keeps the print head inside the print area. I built the receiving half, which used under a fifth of the chip's logic and shows the last command on its seven-segment display.
Illustrative.One message, one G-code line. Each two-character message crosses the Ethernet link, and the RX software adds it to the head's tracked position, clamps it to the 0 to 200 print area, and sends an absolute G1move over UART. Watch theLCcommand hit the left edge and get clamped to X = 0. Set a distance and send your own commands below. The message format, limits, start-up codes, and display layout follow the code in the repository; the command sequence and timing are made up.
Experience and education
- 2022 – present
PhD Researcher, IoT Systems Resilience
Network Enhanced Technologies Lab, Brigham Young University
- 2025
Cybersecurity Research Intern, Subprotocol Discovery
Riverside Research
- 2019 – 2022
Research Assistant, Firmware and GUI Development
Network Enhanced Technologies Lab, Brigham Young University
- Expected April 2027
Ph.D., Electrical and Computer Engineering
Brigham Young University, Provo, UT
- December 2021
B.Sc., Computer Engineering
Brigham Young University, Provo, UT
Teaching
- ECEN 225
Introduction to Computer Systems Laboratory
Instructor (course created by me) · Winter 2023
- ECEN 426
Computer Networks
Teaching assistant · Fall 2022
- ECEN 220
Fundamentals of Digital Systems
Teaching assistant · Fall 2019, Winter 2020
Awards and skills
- 2026
CyberPOWDER Fellowship, University of Utah
- 2024
Best Poster Award, Air Quality: Science for Solutions (Radon Mitigation through Optimized HVAC Scheduling)
- 2023
Best Poster Award, Air Quality: Science for Solutions (Location Monitoring Framework for Citizen Science Sensors)
- 2023
Colosseum Young Gladiators Workshop, Northeastern University
- 2022
First Place, BYU IoT Pi Competition (Seven Segment Sign)
- Python
- C/C++/CUDA
- Linux/Bash
- Computer Networks
- Systems Engineering
- AVR/ESP
- InfluxDB/Grafana
- SystemVerilog/Verilog