Major reorganization of the files

* Set the base version to `v2rev1`
* Rename the project name to `avrhwrng` for simplicity
* Move all old files under `references/`
* Keep v2rev1 source only under the main tree
* Revise readme docs

Makefile

@@ -5,10 +5,10 @@
 ## NOTE: GNU Make only
 
 ## General Flags
-PROJECT = avrhwrng-v2
+PROJECT = avrhwrng
 #MCU = atmega168
 MCU = atmega328p
-TARGET = avrhwrng-v2.elf
+TARGET = avrhwrng.elf
 CC = avr-gcc
 
 ## Options common to compile, link and assembly rules
@@ -21,7 +21,7 @@ CFLAGS += -MD -MP -MT $(*F).o -MF $(@F).d
 
 ## Linker flags
 LDFLAGS = $(COMMON)
-LDFLAGS += -Wl,-Map=avrhwrng-v2.map
+LDFLAGS += -Wl,-Map=avrhwrng.map
 
 ## Intel Hex file production flags
 HEX_FLASH_FLAGS = -R .eeprom -R .fuse -R .lock -R .signature
@@ -31,13 +31,13 @@ HEX_EEPROM_FLAGS += --set-section-flags=.eeprom="alloc,load"
 HEX_EEPROM_FLAGS += --change-section-lma .eeprom=0 --no-change-warnings
 
 ## Objects that must be built in order to link
-OBJECTS = avrhwrng-v2.o
+OBJECTS = avrhwrng.o
 
 ## Objects explicitly added by the user
 LINKONLYOBJECTS = 
 
 ## Build
-all: $(TARGET) avrhwrng-v2.hex avrhwrng-v2.eep avrhwrng-v2.lss size
+all: $(TARGET) avrhwrng.hex avrhwrng.eep avrhwrng.lss size
 
 ## Compile
 
@@ -65,4 +65,4 @@ size: ${TARGET}
 ## Clean target
 .PHONY: clean
 clean:
-	-rm -rf $(OBJECTS) avrhwrng-v2.elf dep/* avrhwrng-v2.hex avrhwrng-v2.eep avrhwrng-v2.lss avrhwrng-v2.map
+	-rm -rf $(OBJECTS) *.o.d avrhwrng.elf dep/* avrhwrng.hex avrhwrng.eep avrhwrng.lss avrhwrng.map

README.md

@@ -3,103 +3,30 @@
 This repository contains a sample code with avr-gcc and the schematics
 for a shield (extended piece of hardware) for Arduino Duemilanove / Arduino UNO.
 
-Note: the code is *not* written in the Arduino programming framework; it
-is written in C and avr-libc at <http://www.nongnu.org/avr-libc/>.
+Note: The code is solely written in C and [avr-libc](http://www.nongnu.org/avr-libc/); no Arduino framework.
 
-## LICENSE
-
-* All AVR source code files are licensed by the MIT License.
-* All schematics files are licensed by CC-BY-4.0.
-
-## Documentation
-
-* See `ipsj-iots2015/` for the set of the documentation (PDF paper in Japanese,
-  slides in English) of my presentation at IPSJ IOTS2015 Symposium on 27-NOV-2015.
-
-## Makefiles
-
-The default `Makefile` is a copy of `Makefile.v2`.
-
-* `Makefile.v1`: for the v1 code
-* `Makefile.v2`: for the v2 code
-* `Makefile.rawtest`: for the rawtest code
-
-## Preliminary test on the rawtest code
-
-26-OCT-2015: removal of filtering from the noise generation circuits conducted.
+## Hardware
 
-By running `avrhwrng-rawtest.c`, all the mixing/filtering code have been
-*removed*, and the AVR outputs random number bytes are just filled in by the
-sampled raw two-bit sequence of PD7 and PD6. No other alteration. 
+The current circuit version is called `v2rev1`. The v2rev1 consists of two
+independent noise generator circuits, and the outputs are amplified to the CMOS
+digital levels, which are connected to ATmega168/328P's input pins (PD6/PD7, or
+Pin 6/7).
 
-The test was conducted on an Arduino Uno R3 with avrhwrng v2rev1 board. The
-sampling rate was 400kHz, and the output rate was 1Mbps. The actual output rate
-measured was ~76kbytes/sec.
+Schematics and photos are available in `schematics/` and `photos/` directories, respectively.
 
-Obviously the output will not pass the statictic tests due to possible skew and
-bias.  Filtering this by SHA512 as a function defined as 
+## Firmware
 
-```
-[512 bytes raw input] + [32 byte output from the previous SHA512 output] -> [64
-bytes of SHA512 output as a part of random number stream]
-```
-
-made the output perfectly usable as a well-whitened random number sequence.
-
-The FIPS 140-2 rngtest result for 1597357500 bytes of output resulted as
-follows (summary: test failure rate: 0.079%):
-
-```
-~/src/rngtest/rngtest -c 638943 < poi1filt.bin
-rngtest 2-unofficial-mt.14
-Copyright (c) 2004 by Henrique de Moraes Holschuh
-This is free software; see the source for copying conditions.  There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
-
-rngtest: starting FIPS tests...
-rngtest: bits received from input: 12778860032
-rngtest: FIPS 140-2 successes: 638440
-rngtest: FIPS 140-2 failures: 503
-rngtest: FIPS 140-2(2001-10-10) Monobit: 62
-rngtest: FIPS 140-2(2001-10-10) Poker: 74
-rngtest: FIPS 140-2(2001-10-10) Runs: 188
-rngtest: FIPS 140-2(2001-10-10) Long run: 180
-rngtest: FIPS 140-2(2001-10-10) Continuous run: 0
-rngtest: input channel speed: (min=762.939; avg=12736.675; max=19073.486)Mibits/s
-rngtest: FIPS tests speed: (min=35.453; avg=126.549; max=129.752)Mibits/s
-rngtest: Program run time: 97340320 microseconds
-```
-
-TestU01 Rabbit and Alphabit tests passed in most cases (~94.8%) on 1Mbit
-samples; 1180 failures detected out of 22750 tests (1 Rabbit, 6 Alphabit tests
-for each sample) performed on 3250 samples.  The longer bit sample tests of
-TestU01 Rabbit and Alphabit tests for the 1992Mbit (249Mbyte) sample were all
-passed.
-
-Result updated 11-MOV-2015: a dieharder result of the 2179368064 byte output
-was very good, only the following three tests showed WEAK, no FAILED test (note
-that the input sequence was rewound as dieharder demanded):
-
-```
-    diehard_rank_6x8|   0|    100000|     100|0.99942491|   WEAK
-        diehard_sums|   0|       100|     100|0.00071977|   WEAK
-      rgb_lagged_sum|  31|   1000000|     100|0.00057416|   WEAK
-```
+* The current version is called `v2rev1`.
+* The v2rev1 gives ~10kbytes/sec of filtered output.
 
-The result poses a fundamental question: *filtering on the hardware RNG board
-is utterly meaningless if a sufficient whitening or
-statictically/cryptographically strong bit mixing is performed on the host
-side.*  Further investigation needed.
+### How the firmware works
 
-## How it works on Version 2
-
-Two independent noise generator circuits, amplified to the CMOS digital levels,
-are connected to ATmega168/328P's input pins (PD6/PD7, or Pin 6/7).
-
-The Timer 0 is set to CTC mode with the closest larger value to the theoretical
-limit, 46 machine cycles or 2.875 microseconds. The theoretical limit of the
-speed is 352000bps or 2.84 microseconds, which is equal to 11000 bytes * 8
-bits/byte * 2 (for von-Neumann test) * 2 (for XORing byte filter). Sampling in
-an equal timing improves the quality of the obtained randomness.
+The Timer 0 of ATmega168/328P is set to CTC mode with the closest larger value
+to the theoretical limit, 46 machine cycles or 2.875 microseconds. The
+theoretical limit of the speed is 352000bps or 2.84 microseconds, which is
+equal to 11000 bytes * 8 bits/byte * 2 (for von-Neumann test) * 2 (for XORing
+byte filter). Sampling in an equal timing improves the quality of the obtained
+randomness.
 
 The MCU runs the following program without any hardware interrupt as an infinite loop:
 
@@ -116,14 +43,6 @@ The MCU runs the following program without any hardware interrupt as an infinite
 Random number stream is obtained through the tty device of Arduino as a binary
 byte stream.
 
-The code runs either on ATmega168 or ATmega328P (only a vector address
-difference).  Tested both on Arduino Duemilanove and Arduino UNO R3 boards.
-
-See `noiseshield-v2/` directory for the schematics (drawn by xcircuit). The
-entire circuits can be built on a standard Arduino prototype board.
-
-## On stream mixing strategy
-
 The current sampling strategy of only applying von Neumann algorithm
 independently for each bit stream and mixing the output into single byte stream
 is chosen for reducing statistical errors.
@@ -138,19 +57,12 @@ Noise gen at PD7 --- von Neumann filter 1
 Noise gen at PD6 --- von Neumann filter 2 
 ```
 
-Treating the two bit streams of PD7 and PD6 as two independent XORing byte
-streams caused the following TestU01 Rabbit test errors:
-
-```
-  1  MultinomialBitsOver              eps
-  2  ClosePairsBitMatch, t = 2      1.9e-68
-  3  ClosePairsBitMatch, t = 4     1.0e-146
-```
+The code runs either on ATmega168 or ATmega328P (only a vector address
+difference).  Tested both on Arduino Duemilanove and Arduino UNO R3 boards.
 
-## Changes for Version 2
+## Documentation
 
-* 25-SEP-2015: Add Timer 0 synchronization code
-* 24-SEP-2015: Initial revision, change PD7/PD6 to digital input, reduce LED blinking from each output bit to each output *byte*, remove interrupt-driven code
+* See `presentations/ipsj-iots2015/` for the set of the documentation (PDF paper in Japanese, slides in English) of my presentation at IPSJ IOTS2015 Symposium on 27-NOV-2015.
 
 ## How to compile
 
@@ -222,36 +134,10 @@ For the tagged binary `v2rev1-20150925`:
 
 * The code is compiled with avr-gcc 4.9.2 running on OS X 10.10.5. Different compilers may generate different binaries.
 * This program is compatible with [optiboot bootloader](https://github.com/Optiboot/optiboot/).
-* PySerial required for the sample Python code. See <http://pythonhosted.org/pyserial/pyserial.html#installation>
-
-## FYI: changes for Version 1
-
-* 23-SEP-2015: Use -O3 altogether
-* 12-AUG-2015: Canceled A0/A1 setup, no connection assumed between AIN1/D7-A1 and AIN0/D6-A0
-* 23-JUL-2015: Performance values revised / *CANCELED*: assumes connection between AIN1/D7-A1 and AIN0/D6-A0 (A0/A1 set to input, no pullup)
-* 19-JUL-2015: Code now targeted for ATmega168 and ATmega328p (modify Makefile: default is for ATmega168).
-* 6-JUL-2015: Code tested under avr-gcc 4.9.2 with HomeBrew build environment.
-* 7-MAR-2009: Code tested under AVR Studio 4.15 with WinAVR-20081205.
-
-## FYI: how it works on Version 1
-
-Two independent noise generator circuits are connected to ATmega168/328P's
-analog comparator inputs (AIN0/AIN1, PD6/PD7, or Pin 6/7). The MCU samples the
-comparator output in a fixed frequency of the Timer 0 interrupts (4
-microseconds = 250kHz).
-
-The sampled output bit stream from AVR analog comparator is filtered through
-von Neumann algorithm of two adjacent sampled bits into a byte stream. Each of
-the two-byte pairs of the byte stream is again filtered into one-byte stream by
-XORing the bytes of the pair with each other.
 
-The code will run either on ATmega168 or ATmega328P.
-
-## FYI: actual output rate for Version 1
+## LICENSE
 
-* Sampling rate: 4 us = 250kHz
-* Output rate for v1rev2: ~2100bytes/sec = 17.6kHz
-* Output rate for v1amp: ~3500bytes/sec = 28kHz
-* Transfer rate from Arduino: 115200bps, 8-bit, no parity raw bytes
+* All AVR source code files are licensed by the MIT License.
+* All schematics files are licensed by CC-BY-4.0.
 
 [End of README.md]