Staff RF and RAN engineer, Denver. Authorized to work in the US without sponsorship.
Eighteen years in wireless, across 5G NR, LTE, Wi-Fi and CBRS. I trace network problems back to their cause, in the field and in the data, and I build the tools and AI systems that make that work faster. Open to capacity, RAN engineering and telecom AI roles.
emir.kursad.ulusoy@gmail.com LinkedIn GitHub CV (PDF) Full portfolio (PDF)
Each piece below has a LinkedIn post that walks through the method step by step. Select a figure to enlarge it.
Capacity and data
Two heatmaps read together. On the demand side, RRC users, timing advance and user throughput show coverage pulling in as load rises, because more throughput needs more PRBs and that lifts the noise floor. On the supply side, PDCCH blocking climbs with user count while PRBs still have headroom, so the scheduler sets the ceiling before the spectrum does.
Capacity and data
Three months of ROP-level data from two cells on the same band. The 5 MHz carrier drops below 1 Mbps at roughly 16 to 34 RRC users; the 15 MHz carrier holds 4 Mbps until about 48 to 66. Each bin is a sample count, so the break point comes straight from measured data rather than a model.
Capacity and data
5G geolocation was stored to about 0.1 m against a GPS accuracy of 5 to 10 m. Re-aggregating on a degree-based grid in AWS Redshift took storage from 180 GB a day to about 2 GB, and the analysis window from three weeks to two years, around 500M samples, small enough to run on a laptop.
AI
Supported max data rate under TS 38.306 for 4x4 MIMO, MCS 28, 19 RBs. One of three models returned 60.94 Mbps, matching my own implementation. The others did the arithmetic fine but recalled the wrong code rate for that MCS. A first check rather than a controlled comparison.
Open-source tool
Signal and quality from nmcli, throughput from curl and iperf3, merged in Python and mapped in QGIS. Used on more than ten indoor deployments in place of a commercial tool at about $5K per engineer.
Read the post on LinkedInData collection on GitHubHeat map reporting on GitHub
Open-source tool
Turns coverage prediction rasters into polygons so geo-filtering runs free in QGIS. Removed about $2K per user in licensing, put several departments on the same prediction workflow, and runs roughly five times faster at national scale.
RF and lab
Mapping-capable analyzers with GPS run into the tens of thousands. A compact setup built for under $6K produced the level map and the frequency plot that placed an external interferer at 668.5 and 682.5 MHz.
Scripted AMOS and Moshell pulls for custom KPIs straight from the node.
Flattens KGET exports into an MO, parameter and value table for bulk diffing against a golden configuration.
A web form request pulls KPI data, draws the charts and returns a formatted report, with nothing installed locally.
US 10,009,784Granted patent
A big data and machine learning method for detecting passive intermodulation from network data instead of a site visit. Built at Yupana.
2024now
Staff Engineer, Wireless, Littleton, CO
Capacity planning for a national Open RAN network of about 300,000 cells, the Python and SQL behind it, and multi-agent AI orchestration for the company platform.
20172024
R&D Director, RF Subject Matter Expert, Walnut Creek, CA
Led 12 engineers through 5G NR FCC and carrier approvals across 250+ sites and clusters. Device and RAN testing, protocol analysis, Open RAN lab work. Director from 2021.
20142017
Senior RF Engineer, Walnut Creek, CA
Built the LTE PIM catcher, granted as a US patent. LTE optimization for venues and events, parameter audits against carrier baselines.
20092014
RF Engineer, Astana and Baku
Ericsson, Nokia and ZTE programs, including a nationwide frequency re-design and a Nokia to ZTE swap.
20082009
RF Engineer, Ankara
RF device R&D and compliance research. Designed and built a 3G repeater.
BS, Electronics Teaching Education, Kocaeli University, top of the department. US evaluation: equivalent to a BS in Electronics Engineering.