FFT IP Core: User Guide

Download PDF
ID 683374
Date 11/06/2017
Public
Document Table of Contents
Document Table of Contents x
1. About This IP Core 2. FFT IP Core Getting Started 3. FFT IP Core Functional Description 4. Block Floating Point Scaling 5. Document Revision History A. FFT IP Core User Guide Document Archive
1. About This IP Core x
1.1. Intel® DSP IP Core Features 1.2. FFT IP Core Features 1.3. General Description 1.4. DSP IP Core Device Family Support 1.5. DSP IP Core Verification 1.6. FFT IP Core Release Information 1.7. Performance and Resource Utilization
1.3. General Description x
1.3.1. Fixed Transform Size FFT 1.3.2. Variable Streaming FFT
2. FFT IP Core Getting Started x
2.1. Installing and Licensing Intel® FPGA IP Cores 2.2. IP Catalog and Parameter Editor 2.3. Generating IP Cores ( Intel® Quartus® Prime Pro Edition) 2.4. Generating IP Cores ( Intel® Quartus® Prime Standard Edition) 2.5. Simulating Intel® FPGA IP Cores 2.6. DSP Builder for Intel® FPGAs Design Flow
2.1. Installing and Licensing Intel® FPGA IP Cores x
2.1.1. Intel® FPGA IP Evaluation Mode 2.1.2. FFT IP Core Intel® FPGA IP Evaluation Mode Timeout Behavior
2.3. Generating IP Cores ( Intel® Quartus® Prime Pro Edition) x
2.3.1. IP Core Generation Output ( Intel® Quartus® Prime Pro Edition)
2.4. Generating IP Cores ( Intel® Quartus® Prime Standard Edition) x
2.4.1. IP Core Generation Output ( Intel® Quartus® Prime Standard Edition)
2.5. Simulating Intel® FPGA IP Cores x
2.5.1. Simulating the Fixed-Transform FFT IP Core in the MATLAB Software 2.5.2. Simulating the Variable Streaming FFT IP Core in the MATLAB Software
3. FFT IP Core Functional Description x
3.1. Fixed Transform FFTs 3.2. Variable Streaming FFTs 3.3. FFT Processor Engines 3.4. I/O Data Flow 3.5. FFT IP Core Parameters 3.6. FFT IP Core Interfaces and Signals
3.2. Variable Streaming FFTs x
3.2.1. Fixed-Point Variable Streaming FFTs 3.2.2. Floating-Point Variable Streaming FFTs
3.3. FFT Processor Engines x
3.3.1. Quad-Output FFT Engine 3.3.2. Single-Output FFT Engine
3.4. I/O Data Flow x
3.4.1. Streaming FFT 3.4.2. Variable Streaming 3.4.3. Buffered Burst 3.4.4. Burst
3.4.1. Streaming FFT x
3.4.1.1. Using the Streaming FFT 3.4.1.2. Changing the Direction on a Block-by-Block Basis 3.4.1.3. Enabling the Streaming FFT
3.4.2. Variable Streaming x
3.4.2.1. Changing Block Size 3.4.2.2. Changing Direction 3.4.2.3. I/O Order 3.4.2.4. Enabling the Variable Streaming FFT 3.4.2.5. Dynamically Changing the FFT Size 3.4.2.6. I/O Order
3.4.3. Buffered Burst x
3.4.3.1. Enabling the Buffered Burst FFT
3.6. FFT IP Core Interfaces and Signals x
3.6.1. Avalon-ST Interfaces in DSP IP Cores 3.6.2. FFT IP Core Avalon-ST Signals
3.6.2. FFT IP Core Avalon-ST Signals x
3.6.2.1. Component Specific Signals
4. Block Floating Point Scaling x
4.1. Possible Exponent Values 4.2. Implementing Scaling 4.3. Unity Gain in an IFFT+FFT Pair
4.2. Implementing Scaling x
4.2.1. Example of Scaling
1. About This IP Core
2. FFT IP Core Getting Started
3. FFT IP Core Functional Description
4. Block Floating Point Scaling
5. Document Revision History
A. FFT IP Core User Guide Document Archive

4. Block Floating Point Scaling

Block-floating-point (BFP) scaling is a trade-off between fixed-point and full floating-point FFTs.

In fixed-point FFTs, the data precision needs to be large enough to adequately represent all intermediate values throughout the transform computation. For large FFT transform sizes, an FFT fixed-point implementation that allows for word growth can make either the data width excessive or can lead to a loss of precision.

Floating-point FFTs represents each number as a mantissa with an individual exponent. The improved precision is offset by demand for increased device resources.

In a block-floating point FFT, all of the values have an independent mantissa but share a common exponent in each data block. Data is input to the FFT function as fixed point complex numbers (even though the exponent is effectively 0, you do not enter an exponent).

The block-floating point FFT ensures full use of the data width within the FFT function and throughout the transform. After every pass through a radix-4 FFT, the data width may grow up to log2 (42) = 2.5 bits. The data scales according to a measure of the block dynamic range on the output of the previous pass. The FFT accumulates the number of shifts and then outputs them as an exponent for the entire block. This shifting ensures that the minimum of least significant bits (LSBs) are discarded prior to the rounding of the post-multiplication output. In effect, the block-floating point representation is as a digital automatic gain control. To yield uniform scaling across successive output blocks, you must scale the FFT function output by the final exponent.

In comparing the block-floating point output of the Altera FFT MegaCore function to the output of a full precision FFT from a tool like MATLAB, you must scale the output by 2 (–exponent_out) to account for the discarded LSBs during the transform.

Unlike an FFT block that uses floating point arithmetic, a block-floating-point FFT block does not provide an input for exponents. Internally, a complex value integer pair is represented with a single scale factor that is typically shared among other complex value integer pairs. After each stage of the FFT, the largest output value is detected and the intermediate result is scaled to improve the precision. The exponent records the number of left or right shifts used to perform the scaling. As a result, the output magnitude relative to the input level is:

output*2-exponent

For example, if exponent = –3, the input samples are shifted right by three bits, and hence the magnitude of the output is output*23.

After every pass through a radix-2 or radix-4 engine in the FFT core, the addition and multiplication operations cause the data bits width to grow. In other words, the total data bits width from the FFT operation grows proportionally to the number of passes. The number of passes of the FFT/IFFT computation depends on the logarithm of the number of points.

A fixed-point FFT needs a huge multiplier and memory block to accommodate the large bit width growth to represent the high dynamic range. Though floating-point is powerful in arithmetic operations, its power comes at the cost of higher design complexity such as a floating-point multiplier and a floating-point adder. BFP arithmetic combines the advantages of floating-point and fixed-point arithmetic. BFP arithmetic offers a better signal-to-noise ratio (SNR) and dynamic range than does floating-point and fixed-point arithmetic with the same number of bits in the hardware implementation.

In a block-floating-point FFT, the radix-2 or radix-4 computation of each pass shares the same hardware, with the data being read from memory, passed through the core engine, and written back to memory. Before entering the next pass, each data sample is shifted right (an operation called "scaling") if there is a carry-out bit from the addition and multiplication operations. The number of bits shifted is based on the difference in bit growth between the data sample and the maximum data sample detected in the previous stage. The maximum bit growth is recorded in the exponent register. Each data sample now shares the same exponent value and data bit width to go to the next core engine. The same core engine can be reused without incurring the expense of a larger engine to accommodate the bit growth.

The output SNR depends on how many bits of right shift occur and at what stages of the radix core computation they occur. In other words, the signal-to-noise ratio is data dependent and you need to know the input signal to compute the SNR.


Section Content
Possible Exponent Values
Implementing Scaling
Unity Gain in an IFFT+FFT Pair

Level Two Title