Massive Audio D2500.1 block series User Manual download pdf (Page 25)

track of the current bit was only 6 bits wide, which supports only operand bit

widths up to 63. We extended the width of this counter to 7 bits to solve this

issue.

ModelSim was also used to verify the functionality of the BCD module. After

uncovering a bug in the percentage of pixels kept during integration, the ﬁrst

area of investigation was the BCD module, since it was the least tested module.

It was trivial to write a test bench for the module and verify the output was

correct using ModelSim.

7.3 Julia implementations/tests of algorithms

Prior to implementing the perspective transforms on the FPGA, we deﬁnitely

wanted to verify the soundness of our approach in software ﬁrst. We settled

on the use of the Julia programming language for this purpose, though in all

likelihood MATLAB or Python(+numpy/scipy/matplotlib) would have served

just as well. This was by far the most eﬀective means of testing whether an al-

gorithm is correct or not, since one can rely on the beneﬁts of fast debug/iterate

cycles in software. This is particularly true in the case of languages featuring a

REPL (read, evaluate, print loop) based interpreter, such as Julia. Correcting

code is as simple as making some changes, re-including the ﬁle in the REPL,

and rerunning. This helped us particularly in the veriﬁcation of the pixels kept

and perspective params module. As an aside, to avoid using too many lan-

guages for software implementations, we used Julia for our code generator for

the accel lut as well. As a concrete example of a serious bug caught using the

help of our software implementations, we were able to track and correct an error

in the computation of p

and p

using this method. Essentially, the bug was

an interchange of two terms in the closed form expressions that occurred while

transcribing the closed form solutions we found for p

and p

from paper to the

computer code.

7.4 Labkit I/O

We wired up the logic analyzer to the accelerometer pins to verify that the acc

module was correctly initializing the accelerometer (e.g. the par to ser module

was producing the correct serial bit stream for the register address and data).

It turned out that the bug in the accelerometer was elsewhere (see the next

section).

7.5 Staring at code/data-sheets

This is deﬁnitely a questionable inclusion, and we do not recommend this

method in general. Nevertheless, it is often eﬀective when done correctly and in

the right spirit, since it forces one to rethink and step through the logic again,

questioning all assumptions. For instance, once we conﬁrmed that the divider

implementation was incorrect, we had to examine the divider code line by line.

1 2 ... 20 21 22 23 24 25 26 27 28 29 30 ... 244 245

Comments to this Manuals

No comments

Massive Audio D2500.1 block series User Manual Page 25

Comments to this Manuals

Related products and manuals for Car audio amplifiers Massive Audio D2500.1 block series