Anomaly detection in a controller area network

1. A computing apparatus, comprising:

a hardware platform;

a network interface to communicatively couple to a bus lacking native support for authentication; and

an anomaly detection engine to operate on the hardware platform and configured to:

receive a first data stream across a first time;

symbolize and approximate the first data stream, comprising computing a first window sum and dividing the data stream into frames;

symbolize at least a portion of each frame, wherein symbolizing comprises dividing a data space of potential values for data units of the frame into a plurality of n bucket symbols, and assigning each data unit a symbol according to its bucket;

compute a first sum across the time, wherein computing the first sum comprises dividing the data stream into a plurality of sequences of length L s symbols, with each sequence overlapping with a previous sequence by L o symbols, and tallying a number of occurrences of each sequence;

receive a second data stream across a second time substantially equal in length to the first time, the second data stream comprising data across a plurality of dimensions from the first data stream;

symbolize and approximate the second data stream, comprising computing a second window sum;

compute a difference between the first window sum and the second window sum;

determine that the difference exceeds a threshold and that a correlation across the plurality of dimensions is broken; and

flag a potential anomaly.

2. The computing apparatus of claim 1 , wherein tallying the number of occurrences of each sequence comprises recording a tally in an n×n×L s array, wherein each cell in the array represents a possible value for a symbol sequence.

3. The computing apparatus of claim 2 , wherein computing the difference between the first window sum and the second window sum comprises computing an absolute value of a cell-wise difference between each cell in the array, and summing the absolute values of the cell-wise differences.

4. The computing apparatus of claim 1 , wherein a data unit is an 8-bit byte, and wherein a portion of each frame is a fixed byte of each frame.

5. The computing apparatus of claim 4 , wherein the fixed byte is a first byte.

6. The computing apparatus of claim 1 , wherein n=8.

7. The computing apparatus of claim 1 , wherein n=4.

8. The computing apparatus of claim 7 , wherein L o =1.

9. The computing apparatus of claim 8 , wherein the first time is two seconds.

10. The computing apparatus of claim 9 , wherein the threshold is 3,000.

11. The computing apparatus of claim 1 , wherein the anomaly detection engine is further to identify the anomaly as a potential intrusion.

12. The computing apparatus of claim 11 , wherein the anomaly detection engine is further to take remedial action, or to notify a security agent to take remedial action.

13. The computing apparatus of claim 1 , wherein the bus is a controller area network (CAN) bus for an automobile.

14. The computing apparatus of claim 13 , wherein the computing apparatus is configured to operate as an in-vehicle network gateway for the CAN bus.

15. One or more tangible, non-transitory computer-readable storage mediums having stored thereon executable instructions to:

communicatively couple a network interface to a bus lacking native support for authentication;

receive a first data stream across a first time, the first data stream comprising data across a plurality of dimensions, wherein data of the plurality of dimensions are correlated with one another;

symbolize and approximate the first data stream, comprising computing a first window sum and dividing the data stream into frames;

symbolize at least a portion of each frame, wherein symbolizing comprises dividing a data space of potential values for data units of the frame into a plurality of n bucket symbols, and assigning each data unit a symbol according to its bucket;

compute a first sum across the time, wherein computing the first sum comprises dividing the data stream into a plurality of sequences of length L s symbols, with each sequence overlapping with a previous sequence by L o symbols, and tallying a number of occurrences of each sequence;

receive a second data stream across a second time substantially equal in length to the first time, the second data stream comprising data across the plurality of dimensions from the first data stream;

symbolize and approximate the second data stream, comprising computing a second window sum;

compute a difference between the first window sum and the second window sum;

determine that the difference exceeds a threshold and that the correlation across the plurality of dimensions is broken; and

flag a potential anomaly.

16. The one or more tangible, non-transitory computer-readable mediums of claim 15 , wherein tallying the number of occurrences of each sequence comprises recording a tally in an n×n×L s array, wherein each cell in the array represents a possible value for a symbol sequence.

17. The one or more tangible, non-transitory computer-readable mediums of claim 15 , wherein computing the difference between the first window sum and the second window sum comprises computing an absolute value of a cell-wise difference between each cell in the array, and summing the absolute values of the cell-wise differences.

18. The one or more tangible, non-transitory computer-readable mediums of claim 15 , wherein a data unit is an 8-bit byte, and wherein a portion of each frame is a fixed byte of each frame.

19. The one or more tangible, non-transitory computer-readable mediums of claim 18 , wherein the fixed byte is a first byte.

20. The one or more tangible, non-transitory computer-readable mediums of claim 15 , wherein n=4 and L o =1.

21. The one or more tangible, non-transitory computer-readable mediums of claim 20 , wherein the first time is two seconds.

22. A computer-implemented method of providing anomaly detection on a bus lacking native support for authentication, comprising:

communicatively couple a network interface to the bus;

receiving a first data stream across a first time, the first data stream comprising data across a plurality of dimensions, wherein data of the plurality of dimensions are correlated with one another;

symbolizing and approximating the first data stream, comprising computing a first window sum and dividing the data stream into frames;

symbolizing at least a portion of each frame, wherein symbolizing comprises dividing a data space of potential values for data units of the frame into a plurality of n bucket symbols, and assigning each data unit a symbol according to its bucket;

computing a first sum across the time, wherein computing the first sum comprises dividing the data stream into a plurality of sequences of length L s symbols, with each sequence overlapping with a previous sequence by L o symbols, and tallying a number of occurrences of each sequence;

receiving a second data stream across a second time substantially equal in length to the first time, the second data stream comprising data across the plurality of dimensions from the first data stream;

symbolizing and approximating the second data stream, comprising computing a second window sum;

computing a difference between the first window sum and the second window sum;

determining that the difference exceeds a threshold and that the correlation across the plurality of dimensions is broken; and

flagging a potential anomaly.

23. The method of claim 22 , wherein tallying the number of occurrences of each sequence comprises recording a tally in an n×n×L s array, wherein each cell in the array represents a possible value for a symbol sequence.

24. The method of claim 23 , wherein computing the difference between the first window sum and the second window sum comprises computing an absolute value of a cell-wise difference between each cell in the array, and summing the absolute values of the cell-wise differences.

25. The method of claim 22 , wherein a data unit is an 8-bit byte, and wherein a portion of each frame is a fixed byte of each frame.