Chapter 14: Stimulus

Prev: Spike Shape Contents Next: Reports

Chapter 14: Stimulus

The cells being simulated from the brain need to receive outside information such as visual, audio, etc. This information would normally come from other parts of the brain, but if these areas are not being simulated, then the stimulus object allows for signals to be injected into the simulation. See Bottom of page for additional information.

Stimulus Definition

Keyword

Value

Description

STIMULUS

N/A

Indicates the beginning of a stimulus definition section

TYPE

name

(string)

The name which will be used to refer to this object

MODE

name

(string)

The mode of the stimulus. Choices are:

CURRENT which declares the stimulus as a current (I)
VOLTAGE which declares the stimulus as a voltage (V)

PATTERN

name

(string)

The type of stimulus: FILE_BASED, INPUT (i.e. over sockets ), SINE, LINEAR, PULSE, NOISE, FILE_BASED_DIRECT

FILENAME

name

(string)

For file based stimulus, the name of source file
For input stimulus, the location of the server (e.g. cortex.cs.unr.edu)

PORT

integer

The socket port used for input stimulus that NCS will receive stimulus data

TIME_INCREMENT

time

(real)

The time value is used in setting up firing windows; it is the amount of real time each timestep corresponds to.

FREQ_ROWS

value

(integer)

Deprecated. The number of frequencies bins (Each frequency will actually be a column in the stimulus file. The reason the keyword is FREQ_ROWS is because originally the frequencies were laid out as rows. The stimulus file has since been rotated 90 degrees so that the number of rows now corresponds to the number of timesteps the stimulus file will cover. ) This has been succeeded by new keyword - FREQ_COLS.

FREQ_COLS

value

(integer)

The number of frequencies bins (Each frequency will actually be a column in the stimulus file.

CELLS_PER_FREQ

value

(integer)

The number of cells assigned to each frequency bin (Recall that the frequencies are the number of columns in the input file)

DYN_RANGE

min	(real)
max	(real)

The minimum and maximum firing rates of the cells given in Hertz

TIMING

name

(string)

values available are:(Not currently used by NCS)

EXACT
URAND
POISSON

SAMESEED

YES/NO

Not currently used by NCS

AMP_START

value

(real)

The starting amplitude of a stimulus

AMP_END

value

(real)

When certain stimulus types finish, they will have this final amplitude value.

PHASE

value

(real)

For SINE stimulus, this specifies the starting angle is an angle other than zero (0) is desired

VERT_TRANS

value

(real)

For SINE stimulus, this is a vertical translation of the sine wave

WIDTH

value

(real)

For PULSE stimulus, the WIDTH determines the length of time the pulse lasts. This is given in secondes, uses FSV to convert to timesteps

TIME_START

values+

(real)

The times when the stimulus should begin. A stimulus may have multiple start and end times.

TIME_END

values+

(real)

The times when the stimulus should cease. A stimulus may have multiple start and end times.

FREQ_START

value

(real)

The starting Frequency of a stimulus. Used to calculate how many pulses will exist which in turn is used to calculate the change in AMP over the course of execution

SEED

value

(integer)

Random Number Generator Seed Value, This value should be a negative integer(ex. -999). This sets the independent seed to be used by the each stimulus object. This seed is used for the noise generator embedded within the stimulus block.

RATE

value

(real)

for NOISE stimulus, mean firing rate of the Poisson noise in Hz.

TAU_NOISE

value

(real)

for NOISE stimulus, mean time constant for correlation decaying(in second). If TAU_NOISE > 0, will generate inhomogeneous Poisson noise, otherwise, homogeneous.

CORREL

value

(real)

for NOISE stimulus, correlation between randomly generated spikes. If CORREL > 0, will generate correlated noise, otherwise uncorrelated.

END_STIMULUS

N/A

Indicates the end of a stimulus definition section

Example

STIMULUS
        TYPE                    Stim_AI1
        MODE                    CURRENT
        PATTERN                 FILE_BASED
        FILENAME                /home/jwk/bigThing/JM/JM_gasbombs_120x160_AI5.stm
        TIME_INCREMENT          0.005
        FREQ_COLS               1
        CELLS_PER_FREQ          64
        DYN_RANGE               1 77.3043
        TIMING                  EXACT
        SAMESEED                NO
        AMP_START               10
        AMP_END                 10
        WIDTH                   0.001
        TIME_START              0.2
        TIME_END                1.8
        FREQ_START              9999
END_STIMULUS

FILE_BASED_DIRECT Example

FILE_BASED_DIRECT like FILE_BASED stimulus takes stimulus data from a file. But instead of having probabilities of firing, FILE_BASED_DIRECT
stimulus file should actually have amplitude value for each time tick of stimulus duration, for each cell. For Eg. shown below,
there should be 200 columns (corresponding to 200 cells) and 50 lines (corresponding to 50 time ticks) in stimulus file.
(FSV * Stimlus duration = 2000 * 0.025 = 50).

STIMULUS
	TYPE			R2_Negstim_E1
	MODE			CURRENT
	PATTERN			FILE_BASED_DIRECT
	FILENAME		/home/milindz/work/brain_model/stimulus/NegRampStim25ms_0.275nA_FSV2k.txt
	FREQ_COLS		200
	CELLS_PER_FREQ		1
	TIMING			EXACT
	TIME_START		0
	TIME_END		0.025
	FREQ_START		999999
END_STIMULUS

NOISE Example

Currently, NOISE stimulus will only work under VOLTAGE mode.  This type of stimulus is used to generate various types of presynaptic trains 
received by one/multi postsynaptic neuron.  Noise stimulus is able to produce (In)homogeneous (un)correlated Poisson spikes according to 
certain correlation functions, by combining different value of TAU_NOISE and CORREL.


The correlation funciton for inhomogeneous NOISE was given as: {a(t)b(t')} = r^2 + r^2*(sigma^2*delta + (1-delta)*Ca*Cb)*exp(-|t-t'|/tau)
r is the average firing rate, sigma^2 is the global variation(sigma = 1, by default), delta = 1(a=b)/0(a!=b).  The interval of time is drawn 
from exponential distribution with mean at TAU_NOISE. For every interal, N+1 random numbers, noise and Xa(a=1,2,...,N), were generated from 
Gaussian distributions with zero mean and standard deviation sigma, respectively. At the start of each interval, the fring rate for synapse 
a was set to: r(a) = r * (1 + (Xa + noise) * correl) (correl != 0); r * (1 + Xa) (correl == 0) and it was held at this value until the 
start of next interval. 

Since the firing rate is indepedent of time, the homogeneous Poisson process generates every sequence of N spikes over a fixed time interval 
with equal probability.  As a result, the correlation function for homogeneous NOISE was given in a different way.  N0 independent 
Poisson-distributed variables are generated and distributed among the N output variables, correlation of N0 and N follows: 
N0 = N + sqrt(CORREL) * (1-N). If CORREL = 1, N0 = 1 (all output are identical as one random variable); If CORREL = 0, N0 = N (all output are 
independent from each other). 
   
STIMULUS
   TYPE                          Stim_AI0
   MODE                          VOLTAGE
   PATTERN                       NOISE
   TIME_START                    0.000
   TIME_END                      1.000
   RATE                          10.0
   TAU_NOISE                     0.020
   CORREL                        0.0
   SEED                         -11
END_STIMULUS

Stimulus Injects

A definined Stimulus may be injected multiple times and to differnt destinations. Each injection of a Stimulus requires a separate Stimlus Inject definition.

Keyword

Value

Description

STIMULUS_INJECT

N/A

Indicates the start of a stimulus inject definition section

TYPE

name

(string)

The name which will be used to refer to this object

STIM_TYPE

name

(string)

The name of the stimulus this injection is for

INJECT

Column	(string)
Layer	(string)
CellType	(string)
Compartment	(string)
Probability	(real)

The destination of the stimulus

END_STIMULUS_INJECT

N/A

Indicates the end of a stimulus inject definition section

Example

STIMULUS_INJECT
        TYPE                    StimInject_AI1
        STIM_TYPE               Stim_AI1
        INJECT                  AI1  Lay4    Exc-cNAC        s1      1
END_STIMULUS_INJECT

Notes
File based Stimulus:

Consider a stimulus input file with, say, 129 columns with 429 elements per column. The 129 columns correspond to 129 Fourier frequency bins. The first samples in each column corresponds to the first Fourier window time, the second samples in each column corresponds to the second time step, and so on. The samples represent arbitrarily scaled energy values.

When ncs reads the file in, it scales this input to probabilities, 0 to 1.0.

CELLS_PER_FREQ gives the number of cells in each frequency bin (column in the input file) in the cortex column. For the example above, if CELLS_PER_FREQ is 10 then the column should have 10 x 129 = 1290 total cells. [If the number of cells in cell group are less than result of CELLS_PER_FREQ x frequency bins, then NCS gives a error message. It is ok if number of cells in cell group is greater than CELLS_PER_FREQ x frequency bins.]

The TIME_INCREMENT specifies how much real time each time step (line in the input file) corresponds to. For audio data Fourier windowed in 2.5 ms chunks, as was the case in earlier runs, the TIME_INCREMENT would be .0025. If TIME_INCREMENT is 0.0025 (2.5 msec), then one data sample (a line) from the stimulus input file is used as a stimulus for that duration. So, if you need a stimulus for 50 milli-seconds, the stimulus input file would have 20 lines (or rows) of probability values.
The TIME_INCREMENT is, or should be, used to back-calculate the spiking probability per time step, as described below.

The DYN_RANGE settings describes the minimum and maximum firing rate of the cells in Hz. 400 is a high value biologically, which corresponds to a 2.5 ms cycle time, but 400 has been used in past runs.

The somewhat tricky calculation for determining the probability of each cell in a CELLS_PER_FREQ bin firing then goes like this. Let's consider first the frequency bin and timestep of maximum energy in the input file. We want to arrange things so that each cell in the CELLS_PER_FREQ bin for that frequency and *for those simulation timesteps within that TIME_INCREMENT time* will fire with a probablity such that the rate of firing will be the maximum value specified in the dynamic range (400 Hz in our examples). For that frequency bin and timestep in the input that has a half-of-maximum energy (scaled probability of .5), we calculate a firing probability for the cells so that during simulation time steps in the TIME_INCREMENT corresponding to that input sample, the cells in the corresponding CELLS_PER_FREQ bin in the column will fire at half the dynamic range maximum.

The AMP_START, AMP_END, and WIDTH describe the pulse that is applied if a cell is firing.

Another File Based Description:

Given a text file,
The columns will represent the different frequencies
< >freq_1 freq_2 ... freq_n

The rows are the data such that 1 data row * time_freq_incr = total time of the stimulus
< >data_row_1 ...
< >data_row_2 ...
< >...
< >data_row_m ...

Example:
A textfile has 400 columns, 100 rows
A simuluation has 1200 cells and wants to inject stimulus every 2.5 ms
To inject all cells:
< >FREQ_COLS 400
< >CELLS_PER_FREQ 3
< >TIME_INCREMENT 0.0025 1.00

Prev: Spike Shape Contents Next: Reports

Last updated Friday 11/2/2007