Streaming Median

Designing a streaming sliding-window median calculator in SystemVerilog

This hard hardware design problem tests whether you can implement a real-time statistical operator on streaming data—a common requirement in financial signal processing and market-data pipelines. You must maintain a fixed-size window, sort or track order dynamically, and emit results every cycle once the window fills, all while respecting AXI-Stream handshake semantics and clock-by-clock timing.

The core challenge is choosing an efficient data structure and sorting strategy. A naive approach (storing 7 elements and fully sorting every cycle) works but may not meet latency or area constraints. Strong solutions consider trade-offs between insertion-order tracking, partial sorting, or specialized median-finding algorithms. You must also handle the stream protocol correctly: buffer incoming data only when i_data_valid is asserted, emit median only after 7 valid elements arrive, and reset state cleanly when i_data_last marks the end of a stream segment.

• Circular buffers and window management
• Sorting networks vs. insertion-sort vs. heap-based selection
• Pipelining and throughput: one result per input cycle after warm-up
• AXI-Stream protocol: valid/ready/last handshaking
• Synchronous reset and edge-case handling (short streams, back-to-back streams)