Altera Stratix GX Transceiver - Manual (Page 208)

8–4 Altera Corporation
Stratix GX Transceiver User Guide January 2005
Pattern Generator
deterministic data dependant components are masked out. However, for
more accurate measurements, use a TIA or some type of jitter separation
software to break down the random and deterministic components.
High-frequency mode is also useful when trying to characterize the
high-frequency losses in the time domain. The delta amplitude difference
between the high-frequency pattern and the low-frequency pattern can
give you a first-order approximation of the high-frequency losses due to
the skin effect and dielectric losses. This method is useful only for a
first-order approximation; use extractions of RLGC values with 2D and
3D field solvers to determine more accurate loss coefficients.
High-frequency mode is enabled when option 2 is selected in the
Quartus II software under what self test mode do you want to use?
Enable the 8b/10b encoder to generate the high-frequency pattern. If it is
disabled, an 8'b10110101 character is sent instead of the
10'b1010101010
character.
Low-Frequency Mode Generator
In low-frequency mode, the BIST generator transmits a K28.7 -/+
character (8'b11111100) into the 8b/10b encoder to generate a
10'b0011111000
or 10'b1100000111 low-frequency character. The
low-frequency data transition toggles at one-tenth the data rate of the
high-frequency pattern.
Like the high-frequency pattern, the low-frequency pattern is DC
balanced with the number of ones equal to the number of zeros. This fact
is important when trying to perform a first order random jitter
measurement. You can measure this jitter using an oscilloscope with a
histogram defined at the zero crossing point. This method is crude, but
still yields a first-order estimated value, because the majority of the
deterministic data-dependant components are masked out. However, for
more accurate measurements, use a TIA or some type of jitter separation
software to break down the random and deterministic components.
Because the data transitions in a slower frequency, the signal is less prone
to high-frequency losses. As a result, the signal is able to rise to a higher
amplitude than the high-frequency components. Therefore, the delta
between the two measurements yields a first order approximation of the
high-frequency losses in the time domain. Once again, this approach is
useful only for a first-order approximation. Use extractions of RLGC
values with 2D and 3D field solvers to determine more accurate loss
coefficients.