68
3 SAMPA Chip Implementation
3.2.4.3 Alternative Lossy Data Compression
For an efficient use of the sparse available bandwidth in daisy-chained mode, it is
important to reduce the amount of data sent per cluster. However, as discussed in
Sect. 2.4.3.3, the implemented cluster sum compression, as seen in Fig. 3.18, is sufficient for the MCH detector and so the following two methods for reduction of the
amount of data sent are only proposed, but they have not been implemented.
Zero suppression with combined cluster sum and length
An alternative to the normal cluster sum mode is to combine the cluster sum and
the cluster length into a 20-bit word. This can be done by constraining the size of a
cluster to 31 samples, which will make it fit into 5 bits, leaving 15 bits for the cluster
sum. The cluster sum can normally not overflow, as 31 · 2
10
< 2
15 , but the cluster
size can possibly overflow, as the maximum size of a cluster is potentially 2
10 if the
zero suppression level is not configured properly. However, clusters can be artificially
broken up into multiple clusters, if the length of the cluster is more than 31. As the
time stamp for the start of a cluster and the length is present in the data, it is possible
to join clusters together again when doing analysis. Since the average length of a
cluster with the present longest shaping time of 300 ns and highest sampling rate of
20 MHz is about 20, the potential for overflow should be low, as long as the zero
suppression level is set correctly. With this method, the total compressed size of a
cluster would go down from four 10-bit words to three in optimal conditions, so the
reduction in payload size would be 25 %.
Zero suppression with variable length sum
In this configuration, the sum of the cluster is kept in a 10-bit word when the sum
is smaller than 2
10 , and in a 20-bit word if larger. This is encoded in one bit of
the length information, and so the allowable length of a cluster will be limited to
9 bits. Optimally this method can reach the same payload reduction of 25 % as the
previously described method, it is less complex to implement, uses less area, and
will not reach a bigger payload size than the normal cluster sum mode, except for
when there are clusters longer than 511, which is uncommon. However, it is quite
sensitive to the average pulse height, and with an average pulse height higher than
20 % of the ADC range there would be little benefit to this method.
3.2.4.4 Huffman Compression
To provide the possibility to transmit lossless data with fewer serial links, Huffman
encoding has been implemented as was specified in Sect. 2.4.3.1. The length-limited
Huffman table is implemented by reusing the constants for the tail cancellation
filter of four channels. As the intention of Huffman is to transmit lossless data,
the tail cancellation should not be enabled. The tail cancellation filter has nine 13-
3 SAMPA Chip Implementation
3.2.4.3 Alternative Lossy Data Compression
For an efficient use of the sparse available bandwidth in daisy-chained mode, it is
important to reduce the amount of data sent per cluster. However, as discussed in
Sect. 2.4.3.3, the implemented cluster sum compression, as seen in Fig. 3.18, is sufficient for the MCH detector and so the following two methods for reduction of the
amount of data sent are only proposed, but they have not been implemented.
Zero suppression with combined cluster sum and length
An alternative to the normal cluster sum mode is to combine the cluster sum and
the cluster length into a 20-bit word. This can be done by constraining the size of a
cluster to 31 samples, which will make it fit into 5 bits, leaving 15 bits for the cluster
sum. The cluster sum can normally not overflow, as 31 · 2
10
< 2
15 , but the cluster
size can possibly overflow, as the maximum size of a cluster is potentially 2
10 if the
zero suppression level is not configured properly. However, clusters can be artificially
broken up into multiple clusters, if the length of the cluster is more than 31. As the
time stamp for the start of a cluster and the length is present in the data, it is possible
to join clusters together again when doing analysis. Since the average length of a
cluster with the present longest shaping time of 300 ns and highest sampling rate of
20 MHz is about 20, the potential for overflow should be low, as long as the zero
suppression level is set correctly. With this method, the total compressed size of a
cluster would go down from four 10-bit words to three in optimal conditions, so the
reduction in payload size would be 25 %.
Zero suppression with variable length sum
In this configuration, the sum of the cluster is kept in a 10-bit word when the sum
is smaller than 2
10 , and in a 20-bit word if larger. This is encoded in one bit of
the length information, and so the allowable length of a cluster will be limited to
9 bits. Optimally this method can reach the same payload reduction of 25 % as the
previously described method, it is less complex to implement, uses less area, and
will not reach a bigger payload size than the normal cluster sum mode, except for
when there are clusters longer than 511, which is uncommon. However, it is quite
sensitive to the average pulse height, and with an average pulse height higher than
20 % of the ADC range there would be little benefit to this method.
3.2.4.4 Huffman Compression
To provide the possibility to transmit lossless data with fewer serial links, Huffman
encoding has been implemented as was specified in Sect. 2.4.3.1. The length-limited
Huffman table is implemented by reusing the constants for the tail cancellation
filter of four channels. As the intention of Huffman is to transmit lossless data,
the tail cancellation should not be enabled. The tail cancellation filter has nine 13-