Home Page > > Details

ECE2072Help With ,Help With Python/c++ Programming

MONASH UNIVERSITY
Department of Electrical and Computer Systems Engineering
ECE2072 ASSIGNMENT
Due Date: Friday 21st October 2022 11:55 pm.
Submission: Online submission using a Moodle assignment. Individual
submissions are required.
Assessment: 10% final mark, marked out of 60. Plagiarism or collusion will
result in zero marks for the originator and the copier.
A marking rubric is provided at the end of this document.
Late Policy: Late submission will incur a penalty of 10% for each day late.
Assignments submitted more than a week late will not be
assessed. Apply for special consideration for late submission due
to documented serious illness/circumstances outlined here:
https://www.monash.edu/exams/changes/special-consideration
Assignment Feedback and Grading
The objective of the assignment is to provide an opportunity to design logic circuits and
give you feedback on your understanding of the design of FSMs and their
implementation in a Hardware Description Language.
A solution that appears to work will not necessarily receive full marks. A good
design is more than a "working solution" - it is clear, uses consistently labelled signals,
is concise, efficient, easily maintained and understood, well documented and states
assumptions where necessary.
Use of high level behavioural Verilog style will be rewarded over low level Boolean
logic expressions and individual flip-flop instantiations. Ensure you understand the
Good Style Guide and Marking Rubric at the end of this document. Students who
have participated and understood the ECE2072 laboratory experiments will have the
skills to complete this assignment.
2
Submission Requirements (via campus specific dropbox):
Please use Moodle to submit the following five files and one video as a zip:
1. A single PDF file: containing your written answers to each part.
○ The questions to be answered are listed under “REPORT Part 1”, etc, at
the end of each section.
○ Each part must start on a new page and must be in the correct order:
Part 1, Part 2, and then Part 3.
○ Verilog code be “pretty printed” in colour - instructions on producing
this will be on Moodle.
2. Three .sof files: your three binary files, forming the solution for each part:
○ They should be labelled MyPart1.sof, MyPart2.sof and MyPart3.sof
3. Video: Part of A video of 3 minutes or less of you demonstrating your code
working on an FPGA
○ The video should have a length of 3-4 minutes
○ You must appear in your video
○ The (programmed) FPGA must appear in your video
○ You should demonstrate the final functionality of your design
(MyPart3.sof) and discuss:
i. your understanding of its functionality
ii. your design methodology
iii. any design decisions you made
4. A final file should contain all your Verilog code named with your student ID
as ID.v.
○ Do not include the modules below the FSM module since these are
not to be changed and will generate false plagiarism alerts.
○ You will not be able to upload this file separately, you must submit it
as a zip, or Moodle will not accept it.
○ If your ID is 12345678 the filename would be 1234568.v - do not
pretty print this file since it must be readable plain text.
○ This file will be checked for plagiarism electronically against all other
code using the Measure of Software Similarity MOSS:
https://theory.stanford.edu/~aiken/moss/
3
Please note that both providers and recipients of any plagiarised reports or parts of
reports will be penalised via Faculty of Engineering disciplinary proceedings –
plagiarism is treated very seriously and can lead to exclusion from the university.
Repeat students should note that this assignment is different to previous years and, in
any case, it is not permitted to submit code submitted previously for assessment.
4
Introduction to the Design Problem
The assignment’s aim is to design a cognition timer that measures the time for a user
to process a request to complete a simple task. Cognition times are important for
assessing our ability process information quickly and accurately. To keep things simple
here we use a basic reading, processing, action planning and implementation task as
follows: A pair of octal digits is displayed, each in the range 0-7. The user adds the
numbers and enters the result as a 4 bit binary number using slide switches SW[3:0] on
the DE2 board. The user indicates the answer is complete by changing the submit slider
switch SW[17] to 1. The time from displaying the two octal digits to the switches are
submitted is then displayed in 100 milliseconds. Should a wrong switch value be
detected after SW[17] is 1 or the user does not respond quickly enough, an error is
shown instead of the time. More details are provided below. To aid your understanding
of the assignment requirements, DemoPart1.sof, DemoPart2.sof and DemoPart3.sof
demonstration files are provided on Moodle for downloading to a DE2 board.
The parts 1 and 2 are intended to be an easy way to familiarise yourself with some of
the basic Verilog modules that have been provided on the Moodle site. Part 3 is more
challenging and requires you to develop a “puppeteer” style finite state machine (FSM)
that controls the whole cognition timer design.
5
Part 1: A Simple Timer Counter and Display
Design a timer counter that provides a running display in units of 100 milliseconds. On
startup or downloading of the .sof file the timer should initially display 0000 and then
increment every 100 milliseconds until it reaches the decimal number composed of the
the last 2 digits of your ID number + 100. For example, if your ID number is 98765432,
your counter timer counts up to 0132 in 13.2 seconds. The counter then wraps back to
0000 and the whole process repeats. Your design should ensure each output value,
including the largest (0132 in our example) persists for 100 milliseconds each.
Start by creating a new Verilog module named MyPart1for your design and instantiate
this within the top level module named assign2022 – ie the module that connects to the
standard board signals such as CLOCK_50, SW[17:0] and KEY[3:0]. Within MyPart1
instantiate the Timer and DisplayTimerError modules and add some logic of your own.
Use SW[17:16] to enter error codes (see the provided module DisplayTimerError),
SW[15] to enable the display, and SW[14] to freeze the output. Make HEX7 and HEX6
display the last two digits of your ID (rather than P1 as in the demo, so we know this is
your demo!). Connect the reset iRst to ~KEY[0]. Test your design. It should be similar
to DemoPart1.sof that counts to 0132 – check it out by downloading to a DE2 board.
REPORT Part1:
● 1.1: Explain how your design has made the time duration spent
displaying your last count value 100 millisecond.
● 1.2: Include pretty printed Verilog just for your assign2022 and
MyPart1 module – do not include other modules. How to “pretty print”
is described on Moodle.
In your submission, you must include a file named MyPart1.sof (rename the file that
Quartus produces for download to the FPGA).
6
Part 2: Display and Generation of Two Octal Digits
At the top level in the module assign2022 instantiate the module
OctNumbersGenerationDisplay that is provided. In part 2 the top level module
assign2022 should display a new pair of numbers, each in the range 0 to 7, on the seven
segment displays HEX5 and HEX4. A new pair is generated on each posedge iClk
when SW[0] is 1. The sum of the two octal numbers defines which switches the user
should set in the final cognition timer design. We are not implementing the checking
for the correct switches here in part 2. That is for part 3. We are only aiming to display
and generate the two octal numbers in this part.
Use SW[5] connected to the port iChooseRandID to choose the source of the 6 bit
number (ie two 3 bit octal digits). A random number generator is the source when
SW[5]=1 and your ID number sequence is used when SW[5]=0. The ID sequence uses
a module called NextIDdualOctal. You need to complete the implementation of
NextIDdualOctal as follows:
● Module NextIDdualOctal has the standard iClk and iRst inputs for the system
clock (usually CLOCK_50) and system reset (usually a synchronised
~KEY[0]). After a reset, followed by a request for the next ID digit iNext=1,
the module output oIDdualOctal should produce, on the next iClk positive edge,
two octal digits. That is two numbers in the range 0 to 7 are produced. The user
will need to add these and enter the sum in binary in Part 3. In this part we
have made the two octal digits depend on your ID number as the next point:
● Your ID number has eight digits. Each digit is used in turn - we move to the
next ID digit when iNext=1 and a posedge iClk occurs. The ID digit is
converted to two octal digits by multiplying by 11 and truncating the result to 6
bits (that is the lowest 6 bits). In our example ID=98765432 the first would be
11*9 = 99 = 7’b1100011, truncated to 6 bits gives 6’b100011 which is displayed
as the two octal digits 4 and 3. The second ID digit in our example should
produce 11*8=88=7’b1011000, truncated to 6 bits is 6’b011000 or 3 and 0 octal
digits in our example.
7
● On each successive positive iClk edge when iNext=1, the next ID digit*11
truncated to 6 bits should be produced on oIDdualOctal. Once the last ID
number is reached, we start from the first again.
● Upon reset, the output is the last ID digit*11 truncated to 6 bits, as a
consequence of the requirement that the first request (after a reset) for an ID
digit corresponds to your first ID digit output. Note also that we would like the
initial state after downloading the .sof file to be the reset state. The
configuration file will, by default, initialise all flip-flops to 0, so the reset state
should be all 0’s.
Test your design by connecting the iClk of the module OctNumbersGenerationDisplay
to debounced ~KEY[1] in the top level module assign2022. A manual clock allows
you to slow down the action of iNewOctNumsReq and iChooseRandID to allow testing.
A file DemoPart2.sof is provided for comparison and has an assumed ID = 98765432.
When downloaded (or after a reset on a posedge clock), the output is initially 11*2=22
or in octal 6’o26. After compiling your design, ensure that no hardware multipliers are
generated by your design, since all multiplications by 11 should be completed before
synthesis by the compiler. Also ensure that you do not synthesise unnecessary flipflops (called registers in Quartus) in your design.
REPORT Part2:
● 2.1: How many states are there in a minimal state diagram for the FSM
generated by a good solution for the NextIDdualOctal module?
● 2.2: How many flip-flops should a minimal solution for module
NextIDdualOctal generate? How many does yours generate?
● 2.3: Include pretty printed Verilog for your assign2022 and NextIDdualOctal
modules – do not include other modules since they are not to be modified.
In your submission, you must include a file named MyPart2.sof (rename the file that
Quartus produces for download to the FPGA).
8
Part 3: The Cognition Timer
In this part the basic modules are put together into a module supplied called
CognitionTimer which you need to instantiate at the top level. The CognitionTimer
design consists of instantiations of the modules: Timer, DisplayTimer,
DualOctalGenerationDisplay and FSM. Part 1 of this assignment has given you a good
understanding of the first two modules and whilst Part 2 the third module. In this part 3
you need to control and monitor the first three modules using the finite state machine
module called FSM that you need to complete yourself. It reads in the user input, keeps
an eye on the time, outputs error codes (if the wrong switches are set or timing out with
no switches set), outputs the cognition time, and displays the two octal numbers being
requested at the appropriate times.
The FSM must be implemented as a Moore finite state machine – that is the outputs
(ie the ports of the FSM module that begin with “o”) must only dependent on the state
and not the inputs. This requirement keeps your design simple and it guarantees that
the outputs of FSM only change shortly after a posedge CLOCK_50. You need to work
out what states are needed as you develop your solution. This will become clearer
when you work through the sequence of steps that the FSM module produces:
0. Reset at start up (iRst=1 or on download of a .sof file to the FPGA). This
is usually where we initialise everything and since the FPGA down loads
0s into registers and flip-flops at the start, it makes good sense to use a
state with a zero coding for the reset state.
1. Initially the display is blank for a fixed time of 2 seconds. This is the time
the user is intensely watching for the appearance of the dual octal numbers
to be added and put on the switches.
2. New dual octal numbers are requested by the FSM and displayed,
prompting the user to respond by setting appropriate switches in SW[3:0]
and then setting the submit switch SW[17].
3. The time from display to submit becoming 1, provided the correct sum of
the octal numbers set, is the cognition delay and is displayed in 0.1 second
units. The display persists until the submit switch SW[17] is reset by the
user.
9
4. If the user sets the wrong switch(es) (error 1) or does not respond within 8
seconds (error 2) an error is displayed instead of the the cognition delay.
In the case of error 2, the submit switch SW[17] needs to be set and then
reset by the user to continue. For error 1, the error is removed when the
submit switch SW[17] is reset by the user.
5. Go back to step 1.
It is good idea to develop the FSM module state by state and test as you add new states.
Note that state transitions (ie from state to the next_state in the template code for FSM
provided) can depend on inputs to the FSM and the current state. Use the two always
blocks provided in the template code for the FSM, where the first block updates the
outputs and next state with a combination logic circuit implemented as a case statement.
Remember a combinational logic function represented by an always block needs to
cover all input combinations to avoid inferred latches as discussed in the week 8
workshop. The second always block is provided and just updates the state from
next_state on the rising clock edge.
You must use CLOCK_50 for the global clock, ~KEY[0] for the global reset, SW[17]
for the submit switch, SW[16] for the iChooseRandID, SW[3:0] for the user entry of
the sum in binary, HEX0-3 for the time/error display, HEX4 and HEX5 for the octal
display of the numbers to be added on the switches. The other HEX outputs can be
used for debugging such as observing the state of the FSM. The LEDR[17:0] should
reflect the value of the SW[17:0] as provided in template code assign2022. A file
DemoPart3.sof is provided for comparison and has an assumed ID = 98765432.
Helpful Hints (and time savers!): The slide switches SW[17:0] are not debounced.
So when you slide a switch from 0 to 1, it can produce a sequence that bounces between
0 and 1 before settling on 1 (eg 01001100011101010101111). Likewise for 1 to 0. This
will need to be considered in your design and testing of the FSM. In practice due to the
mechanical design of switches there is an upper limit on the time between the first 1
and the last 0 in this sequence. You can design the FSM to cope with the switch bounce
with minimal overhead and no extra modules.
10
REPORT Part3:
● 3.1: Include pretty printed Verilog for your FSM module. Ensure that you
document the Verilog code. This must include a description of each state and
the use of parameters with brief state names that describe each state. State S1,
S2 etc are NOT acceptable. Clearly the states need different values and the
number of states will determine how many bits are needed for the state. All your
state bits must be contained in the Verilog signal state in your code.
In your submission, you must include a file named MyPart3.sof (rename the file that
Quartus produces for download to the FPGA).
11
Good Style Guide
Follow these guidelines when writing Verilog:
1. Use [N-1:0] ordering of multi-bit signals unless there is a good reason to use
unconvention ordering.
2. Non blocking assignments <= should be used in posedge clock always blocks.
Each bit in <= assignment can synthesise a flip-flop.
3. Do not mix non blocking and blocking assignments in the same always block.
4. Do not drive the same signal in two different blocks within the same module.
5. Cover all input combinations for defining combinational logic from always
blocks. A default assignment at the top of the block is a good way to do this.
6. Think about whether don’t cares should be used in the default case statements.
7. Use synchronous resets unless there is a good reason to violate timing
constraints with asynchrounous resets.
8. Synchronise all asynchronous inputs to the system clock with two cascaded
flip-flops to allow for metastable settling as discussed in week 11 lectures.
9. Use just one global clock (eg CLOCK_50) with posedge triggering in your
designs. Multiple clocks and posedge with negedge triggering can cause timing
problems. Clock enables (CE) can be used to select lower rate sampling than
the global clock.
10. Ensure you have enough bits on signals connected with an in instantiation. The
compiler will issue a warning message if insufficient bits are provided.
11. Use named associations in port mapping for instantiations with more than 2
ports –
eg Dflipflop dff1( .iClk(clk),
.iRst(rst),
.iD(data),
.oQ(out)
);
rather than rely on order:
DFlipflop dff1(clk, rst, data, out);
12. Check all warning messages after compiling in Quartus and ModelSim. See
10.
12
13. Use i and o as a prefix for port names for inputs and outputs of modules. Use
_n as a suffix for asserted low or negative logic signals eg iRst_n.
14. Use all uppercase for PARAMETER_VALUES, capitalise ModuleName and
lowercase for signals.
15. Avoid pointless parameters such as ONE, TWO, THREE for 1, 2, 3.
16. Beware of using operators % / * (modulus divide and multiply) because they
synthesise to large circuits!
17. Check your synthesised circuit after compiling. See Quartus resource usage
and Quartus:Tools-> netlist viewers ->RTL viewer.
18. Use an appropriate radix for readability – eg 4’d10 is the same as 4’b1010 but
the former is usually more readable but this can depend on the context. Decimal
is preferred unless individual bits are needed. This applies particularly to
displaying signals in simulations.
19. Use high level behavioural design style where possible and avoid bit level
instantiation of flip-flops and hand optimised logic gates. For example a multibit counter can be defined using addition as follows
always (posedge clk)
count <= count + 1’b1;
rather than working with individual flip-flop instantiations and Boolean
expressions for each D input.

Contact Us - Email:99515681@qq.com    WeChat:codinghelp
Programming Assignment Help!