Identify binary state of data set (frequency on/off)

I have a large data set that has values ranging from [-3,3] and I'm using a hard limit at 0 as the boundary.

The data has a binary value of 1 when its oscillating from -3,3 at a 56kHz frequency. What this means is that the data will be changing from -3 to 3 and back every N data values where N is typically < 20.

The data has a binary value of 0 when the data is 3 constantly (this can typically last 400+ samples long)

I cant seem to group the data into their binary categories and also know how many samples wide the group is.

Example data:

1.84    |
2.96    |
2.8     |
3.12    |
.       |  I want this to be grouped as a 0
.       |
3.11    |_____
-3.42   |
-2.45   |
-1.49   |
3.12    |
2.99    |  I want this to be grouped as a 1
1.97    |
-1.11   |
-2.33   |
.       |  
.       |  Keeps going until for N cycles

The cycles in-between the logic HIGH state are typically small (<20 samples).

The code I have so far:

state = "X"
for i in range(0, len(data['input'])):    
    currentBinaryState = inputBinaryState(data['input'][i]); # Returns -3 or +3 appropriately

    if(currentBinaryState != previousBinaryState):

        # A cycle is very unlikely to last more than 250 samples
        if y > 250 and currentBinaryState == "LOW": # Been low for a long time
            if state == "_high":
                groupedData['input'].append( ("HIGH", x) )
                x = 0

            state = "_low"

        else:
            # Is on carrier wave (logic 1)
            if state == "_low":
                # Just finished low
                groupedData['input'].append( ("LOW", x) )
                x = 0

            state = "_high"


        y = 0

Obviously, the result isn't as I should expect as the LOW group is very small.

[('HIGH', 600), ('LOW', 8), ('HIGH', 1168), ('LOW', 9), ('HIGH', 1168), ('LOW', 8), ('HIGH', 1168), ('LOW', 8), ('HIGH', 1168), ('LOW', 9), ('HIGH', 1168), ('LOW', 8), ('HIGH', 1168), ('LOW', 8), ('HIGH', 1168), ('LOW', 9)]

I understand I could of asked this on the signal processing SA but I deemed this problem to be more programming oriented. I hope I explained the problem sufficiently, if there's any questions just ask. Thanks.

---

Here is a link to the actual sample data:

https://drive.google.com/folderview?id=0ByJDNIfaTeEfemVjSU9hNkNpQ3c&usp=sharing

Visually, it is very clear where the boundaries of the data lie.

---

Update 1

I've updated my code to be more legible as single letter variables isn't helping with my sanity.

previousBinaryState = "X"
x = 0
sinceLastChange = 0
previousGroup = inputBinaryState(data['input'][0])
lengthAssert = 0
for i in range(0, len(data['input'])):    
    currentBinaryState = inputBinaryState(data['input'][i]);

    if(currentBinaryState != previousBinaryState): # Changed from -3 -> +3 or +3 -> -3 

        #print sinceLastChange

        if sinceLastChange > 250 and previousGroup == "HIGH" and currentBinaryState == "LOW": # Finished LOW group
            groupedData['input'].append( ("LOW", x) )
            lengthAssert += x
            x = 0
            previousGroup = "LOW"

        elif sinceLastChange > 20 and previousGroup == "LOW": # Finished HIGH group
            groupedData['input'].append( ("HIGH", x) )
            lengthAssert += x
            x = 0
            previousGroup = "HIGH"

        sinceLastChange = 0

    else:
        sinceLastChange += 1

    previousBinaryState = currentBinaryState
    x += 1  

Which, for the sample data, outputs:

8
7
8
7
7
596   <- Clearly a LOW group
7
8
7
8
7
7
8
7
8
7
7
8
7
8
7
7
8
7
8
.
.
.

Problem is the HIGH group is lasting longer than it should be:

[('HIGH', 600), ('LOW', 1176), ('HIGH', 1177), ('LOW', 1176), ('HIGH', 1176), ('LOW', 1177), ('HIGH', 1176), ('LOW', 1176)]

• There are only 8 groups made but the plot clearly shows a lot more. The groups appear to be twice the size of what they should be.

I've finally found a solution. I spent far too long getting my head around, what appears to be, a fairly simple problem but it works now.

It won't pick up the last group in the data set but that's fine.

previousBinaryState = "X"
x = 0
sinceLastChange = 0
previousGroup = inputBinaryState(data['input'][0])
lengthAssert = 0
for i in range(0, len(data['input'])):    
    currentBinaryState = inputBinaryState(data['input'][i]);

    if(currentBinaryState != previousBinaryState): # Changed from -3 -> +3 or +3 -> -3 

        #print sinceLastChange

        if sinceLastChange > 250 and previousGroup == "HIGH" and currentBinaryState == "LOW": # Finished LOW group
            groupedData['input'].append( ("LOW", x) )
            lengthAssert += x
            x = 0
            previousGroup = "LOW"

        sinceLastChange = 0

    else:
        if sinceLastChange > 20 and previousGroup == "LOW":
            groupedData['input'].append( ("HIGH", x) )
            lengthAssert += x
            x = 0
            previousGroup = "HIGH"
            sinceLastChange = 0

        sinceLastChange += 1

    previousBinaryState = currentBinaryState
    x += 1           

20 is the maximum number of cycles in the HIGH state and 250 is the maximum number of samples for which the group is in the LOW state.

[('HIGH', 25), ('LOW', 575), ('HIGH', 602), ('LOW', 574), ('HIGH', 602), ('LOW', 575), ('HIGH', 601), ('LOW', 575), ('HIGH', 602), ('LOW', 574), ('HIGH', 602), ('LOW', 575), ('HIGH', 601), ('LOW', 575), ('HIGH', 602), ('LOW', 574)]

When comparing that to the graph and the actual data, it appears to be correct.