Main Content

Implement Hardware-Efficient Real Partial-Systolic Q-less QR with Forgetting Factor

This example shows how to use the hardware-efficient Real Partial-Systolic Q-less QR Decomposition with Forgetting Factor block.

Q-less QR Decomposition with Forgetting Factor

The Real Partial-Systolic Q-less QR Decomposition with Forgetting Factor block implements the following recursion to compute the upper-triangular factor R of continuously streaming n-by-1 row vectors A(k,:) using forgetting factor $\alpha$. It is as if matrix A is infinitely tall. The forgetting factor in the range $0<\alpha<1$ keeps it from integrating without bound.

$$&#xA;\begin{array}{rcl}&#xA;R_0 &#38;=&#38; \mbox{zeros}(n,n)\\[2ex]&#xA;[\sim\;,\;R_1] &#38;=&#38; \mbox{qr}\left(\left[\begin{array}{c}R_0\\&#xA;A(1,:)\end{array}\right],\; 0\right)\\&#xA;R_1 &#38;=&#38; \alpha R_1\\[4ex]&#xA;[\sim\;,\;R_2] &#38;=&#38; \mbox{qr}\left(\left[\begin{array}{c}R_1\\&#xA;A(2,:)\end{array}\right],\; 0\right)\\&#xA;R_2 &#38;=&#38; \alpha R_2\\[4ex]&#xA;\vdots\\[4ex]&#xA;[\sim\;,\;R_k] &#38;=&#38; \mbox{qr}\left(\left[\begin{array}{c}R_{k-1}\\&#xA;A(k,:)\end{array}\right],\; 0\right)\\&#xA;R_k &#38;=&#38; \alpha R_k\\[4ex]&#xA;\vdots\\[4ex]&#xA;\end{array}&#xA;$$

AMBA AXI Handshaking Process

The Data Handler subsystem in this model takes real matrix A as input. It sends rows of A to the QR Decomposition block using the AMBA AXI handshake protocol. The validIn signal indicates when data is available. The ready signal indicates that the block can accept the data. Transfer of data occurs only when both the validIn and ready signals are high. You can set delay for the feeding in rows of A in the Data Handler to emulate the processing time of the upstream block. validOut signal of the Data Handler remain high when rowDelay is set to 0 because this indicates the Data Handler always has data available.

Define System Parameters

n is the length of the row vectors A(k,:) and the number of rows and columns in R.

n = 5;

m is the effective number of rows of A to integrate over.

m = 100;

Use the fixed.forgettingFactor function to compute the forgetting factor as a function of the number of rows that you are integrating over.

forgettingFactor = fixed.forgettingFactor(m)
forgettingFactor =


precisionBits defines the number of bits of precision required for the QR Decomposition. Set this value according to system requirements.

precisionBits = 24;

In this example, real-valued matrix A is constructed such that the magnitude of its elements are less than or equal to one. Your own system requirements will define what those values are. If you don't know what they are, and A is a fixed-point input to the system, then you can use the upperbound function to determine the upper bounds of the fixed-point types of A.

max_abs_A is an upper bound on the maximum magnitude element of A.

max_abs_A = 1;

Select Fixed-Point Types

Use the fixed.qlessqrFixedpointTypes function to compute fixed-point types.

T = fixed.qlessqrFixedpointTypes(m,max_abs_A,precisionBits)
T = 

  struct with fields:

    A: [0×0]

T.A is the fixed-point type computed for transforming A to R in-place so that it does not overflow.

ans = 


          DataTypeMode: Fixed-point: binary point scaling
            Signedness: Signed
            WordLength: 31
        FractionLength: 24

Define Simulation Parameters

Create random matrix A to contain a specified number of inputs.

numInputs is the number of input rows A(k,:) for this example.

numInputs = 500;
A = fixed.example.realUniformRandomArray(-1,1,numInputs,n);

Cast the inputs to the types determined by fixed.qlessqrFixedpointTypes.

A = cast(A,'like',T.A);

Cast the forgetting factor to a fixed-point type with the same word length as A and best-precision scaling.

forgettingFactor = fi(forgettingFactor,1,T.A.WordLength);

Set delay of clock cycles between feeding in rows of A

rowDelay = 1;

Select a stop time for the simulation that is long enough to process all the inputs from A.

stopTime = 2*numInputs*T.A.WordLength;

Open the Model

model = 'RealPartialSystolicQlessQRForgettingFactorModel';

Set Variables in the Model Workspace

Use the helper function setModelWorkspace to add the variables defined above to the model workspace.


Simulate the Model

out = sim(model);

Verify the Accuracy of the Output

Define matrix $A_k$ as follows

$$A_k = \left[\begin{array}{cccc}\alpha^k \\&#38; \alpha^{k-1} \\&#xA;&#38; &#38; \ddots\\&#38;&#xA;&#38; &#38; \alpha\end{array}\right]A(1:k,\; :).$$

Then using the formula for the computation of the $k$ th output $R_k$, and the fact that $[Q,R]=\mbox{qr}(A,0) \Rightarrow A'A = R'Q'QR = R'R$, you can show that

$$A_k'A_k = R_k'R_k.$$

So to verify the output, the difference between $A_k'A_k$ and $R_k'R_k$ should be small.

Choose the last output of the simulation.

R = double(out.R(:,:,end))
R =

    5.2717    0.1389   -0.2004    0.4152   -0.0565
         0    5.5798    0.0561   -0.1367   -0.5387
         0         0    5.4077   -0.0043   -0.1051
         0         0         0    5.6626   -0.1625
         0         0         0         0    5.4957

Verify that R is upper triangular.

ans =



Verify that the diagonal is greater than or equal to zero.

ans =


Synchronize the last output R with the input by finding the number of inputs that produced it.

A = double(A);
alpha = double(forgettingFactor);
relative_errors = nan(1,n);
for k = 1:numInputs
    A_k = alpha.^(k:-1:1)' .* A(1:k,:);
    relative_errors(k) = norm(A_k'*A_k - R'*R)/norm(A_k'*A_k);

k is the number of inputs A(k,:) that produced the last R.

k = find(relative_errors==min(relative_errors),1,'last')
k =


Verify that

$$A_k'A_k = R_k'R_k$$

with a small relative error.

A_k =  alpha.^(k:-1:1)' .* A(1:k,:);
relative_error = norm(A_k'*A_k - R'*R)/norm(A_k'*A_k)
relative_error =


Suppress mlint warnings in this file.


See Also