HLS_FPGA

🚀 Real-Time High-Level Synthesis (HLS) Projects for UltraScale+ FPGAs. Accelerated design pipelines, dataflow architectures, and low-latency compute cores — powered by Vitis HLS, AXI4-Stream, and signal processing for 5G, SDR, and HFT applications. 🔧🔬💡

🔥 HLS_FPGA: High-Level Synthesis Design Vault

Welcome to the HLS_FPGA repository — a cutting-edge collection of real-world, production-grade FPGA accelerators built with Vitis HLS, tailored for:

• 🛰️ Software-Defined Radio (SDR)
• 📡 5G / O-RAN Signal Processing
• 📈 High-Frequency Trading (HFT)
• 🧠 AI/ML Edge Inference
• 🔐 Embedded Security Accelerators

🚀 Key Features

• Fully pipelined, low-latency compute chains (filter → gain → envelope)
• AXI4-Stream + AXI-MM optimized for UltraScale+ SoCs
• Complete Vivado-ready project structure
• Benchmark reports: latency, throughput, utilization
• 🧪 Co-Sim & Waveform validation included

🧠 Concepts Covered

• #pragma HLS DATAFLOW, PIPELINE, UNROLL
• hls::stream, ap_fixed, hls::vector
• Loop optimization, interface pragmas, and dataflow chaining
• AXI4-Lite control interface + AXI-MM streaming buffer logic

💡 Use Cases

• Real-time DSP on FPGAs
• Hardware-accelerated quant finance
• Latency-critical radio and vision systems

🤝 Contribute

Want to optimize a pipeline, add a new design, or share your performance results? Fork this repo and open a PR. Let’s redefine hardware acceleration together.

---

📣 Let's make FPGAs mainstream for developers and innovators everywhere. Star ⭐ the repo, share with your network, and ignite the hardware revolution!

🚨 BREAKING: FPGA-Accelerated DSP Pipelines ⚡📈

Just completed — a fully pipelined, real-time DSP accelerator optimized using Vitis HLS for High-Frequency Trading, signal intelligence, and low-latency inferencing.

🧠 Project: dsp_pipeline — HLS + AXI4-Stream + Dataflow 🔥

---

🚀 Why This Is Mind-Blowing

✅ Dataflow Rocket: Every block—load_process, fir_filter, gain_adjust, store_process—runs truly in parallel. No bubbles. Ultrafast throughput.
✅ Timing Dominated: Required: 10ns. Achieved: 4.641ns. That’s 216+ MHz Fmax without even trying.
✅ DSP-Efficient: Just 18 DSPs, 0 BRAMs, 0 URAMs. UltraLow-resource. UltraFast.
✅ Latency Optimized: Full processing latency = 1,052 cycles (under 10μs @ 100MHz)
✅ Waveform Proven: Cosim ✅. RTL trace ✅. FIFO depth ✅. Parallel timeline ✅. Every bit validated.

---

💥 Why It Matters for HFT

When nanoseconds = millions, a pipelined FPGA-based DSP engine like this can preprocess, filter, amplify, and route market signals before a single CPU interrupt.
No latency tax. No wasted cycles.

---

🔁 Use Cases

• Market signal denoising before tick injection
• Real-time RF front-ends (e.g., GNSS, O-RAN, EW)
• Quant pipelines for proprietary strategy compute
• On-FPGA ML feature pre-processing (waveforms → vectors)

---

📸 Visual Validation Gallery

Report	Preview
🔧 Post-Implementation Resource Usage
⏱️ Co-Simulation Timeline Trace
🔁 Waveform Trace
🧠 Code Analyzer
🧮 Schedule Viewer
📊 Synthesis Report
📟 C Simulation Output

🚀 Lock-Free FPGA Queue for High-Frequency Trading (HFT)

This is not your average FIFO. This is a fully pipelined, BRAM-backed, single-producer single-consumer (SPSC) queue designed to obliterate lock-based latency in high-frequency trading systems.

---

🔥 Why It Matters

When nanoseconds = millions, software locks just don’t cut it. This lock-free SPSC queue:

• ❌ No locks
• ❌ No FIFOs
• ❌ No DSPs
• ✅ Fully pipelined
• ✅ 100% deterministic timing
• ✅ BRAM circular buffer
• ✅ Synthesized with II=1, latency = 5 cycles
• ✅ CoSim, CSim, and Waveform ✅

---

💡 Built For:

• FIX/ITCH packet pipelines
• Market snapshot replay
• FPGA risk-aware strategy engines
• Predictable FIFO-less queuing across accelerators

---

📊 Performance Snapshot

Metric	Value
CoSim	✅ Passed
II	1
Latency	5 cycles
FFs	38
LUTs	175
BRAM	1
URAM	0
DSPs	0

---

🧪 Visual Proof

Preview	Description
	AXI4 + BRAM Block Design
	RTL Cosim Waveform
	RTL Cosim
	Vivado Synthesis
	Real Trade Loop Output

---

🙌 Contribute

Ideas? Want to scale this to MPMC or adaptive queues? PRs welcome.

---

🙌 Contribute

Pull requests, forks, and optimizations welcome.

🚀 HLS-based 5G FIR Chain Accelerator

📈 Ultra-efficient, pipelined, and zero-stall FIR+Gain+Decimation DSP Engine built with Vitis HLS

🔥 Project Summary

Imagine streaming 1024 samples through a FIR → Gain → Decimation pipeline in ~5µs on FPGA — no kernel, no interrupts, no stalls.
We built it. It’s fast, deterministic, and verified.

This HLS project demonstrates a fully pipelined digital signal processing accelerator optimized for 5G PHY, radar DSP, and ultra-low-latency financial compute (HFT).

---

⛓️ Processing Pipeline

load_input() ↓ fir_filter() ↓ gain_control() ↓ decimator() ↓ store_output()

✅ Built with #pragma HLS DATAFLOW and hls::stream for full task-level parallelism
✅ AXI4-Stream interfaces, scalable, synthesizable
✅ UltraScale+ friendly (ZCU104/ZCU9EG)
✅ Bit-accurate waveform and timing validated

---

💡 Technical Highlights

Feature	Status
⏱ FIR Latency	~1,052 cycles
🚀 Max Frequency (Fmax)	> 200 MHz
🧠 DSP Slice Usage	18 DSPs only
📦 BRAM / URAM Usage	Zero
📉 Decimation Factor	2×
✅ Timing Closure	✔️ Clean
🔬 Waveform Match	✔️ Co-sim + RTL

---

📸 Waveform & Benchmark Snapshots

Co-Simulation	Cycle-Accurate	Synthesis Report

Latency Timeline	Vivado Block Design	Vitis HLS Design Implementation

Benchmark

---

🔍 Quick Analysis

Metric	Value
🔄 Full Pipeline Latency	1,052 cycles
🧠 DSP Usage	17 DSPs
🎯 Fmax Achieved (Post-Impl)	> 214 MHz (4.671ns period)
⚙️ Timing Closure	✅ Achieved
🧩 Slices / LUT / FF	0 / 3431 / 4256

---

🎯 Use Cases

• 📡 5G Uplink/Downlink PHY Chains
• 🛰️ Radar Signal Envelope Filtering
• 💸 HFT Market Signal Preprocessing
• 🤖 Edge AI Signal Cleaning (Pre-ML)

---

🧪 How to Run Locally

1. Open in Vitis HLS 2023.2 or later

2. Import source files

3. Run C Simulation

4. Run C/RTL Co-Simulation

5. Check reports for Fmax and utilization

🚀 HFT_Volatility_Engine

A Real-Time Rolling Volatility Engine Designed for High-Frequency Trading — in Hardware.

This project computes rolling variance of tick price streams using a fully pipelined architecture implemented in Vitis HLS with AXI4-Stream interfaces. It mimics real-world financial models used in HFT, portfolio risk estimation, and embedded analytics engines — but executes entirely in silicon.

---

💡 Key Features

• ✅ Zero software latency — pure hardware acceleration
• ✅ Fully pipelined (II = 1) for deterministic throughput
• ✅ Real-time streaming via hls::stream<tick_t>
• ✅ Fixed-point arithmetic with ap_fixed
• ✅ Modular stages: fork_input, preload_window, compute_variance
• ✅ Validated via waveform, CSim, CoSim, synthesis, and timing

---

📈 Rolling Variance Formula

σ² = (Σxᵢ²)/W − (Σxᵢ/W)²

Where:

• W = rolling window size
• xᵢ = tick price at time i

---

⚙️ How It Works

1. fork_input() splits tick input into two streaming branches
2. preload_window() initializes window sums for mean and variance
3. compute_variance() maintains a sliding window using shift registers, computing:

All modules are connected using hls::stream and operate concurrently under #pragma HLS DATAFLOW.

---

📊 Performance Summary

Metric	Value
Latency	6.079 ns
Fmax	151.72 MHz
II	1
DSPs	16
LUTs	8,587
FFs	9,693
BRAMs	6

---

📸 Waveform & Benchmark Snapshots

Co-Simulation Output	Synthesis Utilization	Latency Timeline	Testbench Waveform 1	Testbench Waveform 2

💬 Real-World Use Cases

• ⚡ FPGA-native volatility filters in high-frequency trading
• 📉 DeFi and crypto risk analytics with ultra-low latency
• 🧠 AI models fed by hardware-accelerated financial stats
• 🚀 Edge inference on risk/volatility for embedded finance

---

🤝 Contributing

Pull requests are welcome.
Feel free to fork and enhance — new features, new financial indicators, AXI-MM integration, or CI automation.

---

👤 Author

John Bagshaw
Award-winning researcher and HLS/FPGA/C++ Engineer
🔗 LinkedIn

DM me if you're hiring, collaborating, or building FPGA-based quant infrastructure.

---

⭐ Support

If this project inspires you:

• ⭐ Star the repo
• 🔁 Share with fellow quant or hardware devs
• 💬 Comment your ideas or feature requests

---

Let’s build finance at the speed of light. 🚀
Silicon-native volatility is here.

🚀 MAC_Accelerator_HLS

Real-Time Multiply-Accumulate Hardware Accelerator built using Vitis HLS for DSP/AI workloads.
🧠 Pipelined. 💡 Coefficient-Instantiated. ⚡ Ultra-low-latency.
🎯 Ideal for real-time 5G PHY, radar pipelines, CNN inference, and embedded vision systems.

---

🔧 Architecture Overview

AXI4-Stream Input ↓ load_input() ↓ multiply_accumulate() ↓ store_result() ↓ AXI4-Stream Output

---

📦 Features

• ✅ Fully Pipelined Top-Level MAC
• ✅ Function Instantiation for coefficient-based specialization
• ✅ AXI4-Stream I/O + AXI-Lite Control (Vivado/Vitis Ready)
• ✅ II = 1 (Initiation Interval of 1) for Real-Time Performance
• ✅ Waveform, Cosim, Synthesis Snapshots Included

---

🧠 Core Functions

// Reads two operands from AXI4-Stream

• void load_input(hls::stream<axis_t>& in, data_t& a, data_t& b);

// Computes (a + b) * coefficient

• acc_t multiply_accumulate(data_t a, data_t b, data_t coefficient);

// Streams result to AXI4-Stream

• void store_result(hls::stream<axis_t>& out, acc_t result);

• 🔁 mac_top(...) wires it all into a fully pipelined, real-time MAC unit.

🧠 Core Functions

---

🧪 Testbench Output

MAC result: 90

---

📸 Visual Highlights

3-Panel View

Co-Simulation Waveform	Throughput Timeline	Synthesis Report

2-Panel View

Implementation Block Diagram	Control Analysis

---

📂 Files Included

File Name	Description
HLS_MAC_accel.hpp	Top-level function and type declarations
HLS_MAC_accel.cpp	Pipeline logic with function directives
HLS_MAC_accel_tb.cpp	AXI stream testbench simulation

---

🔗 Interface Summary

Port Name	Description
input_stream_operands	AXI4-Stream input (two operands)
output_stream_result	AXI4-Stream output (MAC result)
coefficient	AXI4-Lite control register for MAC weight
ctrl_axi_lite	AXI4-Lite interface
mac_clk	Clock signal
mac_reset_n	Active-low reset
mac_done_interrupt	Signals when result is valid

---

💡 Real-World Applications

• 📶 5G Baseband DSP
• 🧠 Edge AI Inference Engines
• 🛰️ Satellite Imaging & Radar
• 💻 Embedded Vision SoCs
• 📊 High-Frequency Financial Analytics

---

🤝 Contributing

Want to port this to Zynq, add burst DMA, or wrap it in a full CNN accelerator?
Fork, enhance, and tag me!

---

🔥 Let’s Go Viral Together

If you're building ultra-low-latency pipelines, this project is your launchpad.
⭐ Star this repo and share your MAC benchmarks.