返回列表 发帖

[测试设备] BERTScope 研讨会归来

BERTScope 的研讨会下午1点半开始,由于同时Ansoft在举办《电磁兼容设计》研讨会,于是早上快11点到了会场,完整的听取了华为EMC实验室的人讲解EMC设计规范。之后李宝龙接着讲高速数模混合设计,没讲完,就到了吃饭的时候。记住了几点:

1. 数字信号的干扰源,表现为高幅度离散频谱,幅度随着频率升高迅速衰落。这个讲义中只是简单提了一下,但ppt则拿出典型的干扰源频谱照片给大家给,这跟很多研讨会一样,可以吸引眼球,不至于你一拿到讲义,人就溜啦。
2.数字电源噪声,表现为低幅度连续频谱,从直流一直到无限高频。简单说表现为类“白噪声”特性。
3. 提高隔离度的同时,保持信号完整性和电源完整性之间的平衡。这个应该是重点,可惜讲的非常少。

在吃饭的空间,我问QuickEye/VerifEye 误码率外插将1E-21变成1E-300的意义,答曰这是前瞻性考虑。事实上现在大部分误码率要求在1E-12,而我们面临的仿真问题是,传统工具如Hspice无法进行这么长码型的仿真,于是我就问这个是类似于BERTScan的预估,还是实测,答曰:QuickEye是实测,VerifEye是预估。再问QuickEye 1E-12误码率的仿真速度,答曰:据IBM使用,10Gbps的信号,几个小时可以跑完1E-12的误码率仿真,并且跟BERT比对过,吻合度很高。但Ansoft自己没有进行此类实验。

好厉害,如果QuickEye真能实际跑完1E-12误码率测试,而且准确度值得信赖,那毫无疑问,将会成为高速码型的仿真利器,甚至可以弥补示波器测量深度太浅的缺陷。不知有没有人实际用过?

接着下午就到了BERTScope的会场,一帮台湾人和1个黑人做技术支持,Tek在旁协助。

1. 弄清楚了Jitter Map的分类,在下面这张图上,DJ的细分只有DDJ和BUJ,而没有通常的PJ,原来BertScope认为PJ是BUJ的一部分。而BUJ是同步码型的低通滤波产生的,help文件中有详细描述,我还没有完全弄明白。


2. jitter injection的功能的确方便,当日展示的就是USB 3.0的测试。可以调用一致性模板,进行自动的jitter tolerance测量。也可以手动调节诸如SJ的幅度,找出误码产生时的抖动容限。还可以方便的进入loop back模式,在CP0和SKP码型之间切换。Symwave的兄弟拿了他们公司的U盘和硬盘过来测试。刚开始好一阵子,都进不了loop back,原来是Device 没有DC Block,在码型输出端口加了DC Block后,终于顺利的进入了回环。但是利用模板测试,总是不能扫除jitter tolerance,原来在500kHz SJ 默认幅度是 200%UI,而device总是失锁。于是改为手动,降低SJ的幅度,终于可以进行。但也暴露出BertScope的一个问题,就是不能从小到大的诸如抖动,一旦jitter tolerance超限,则会失锁,无法再继续往下测试,从而找出真实的jitter tolerance曲线。


3. BERTScope的均衡功能只有CTLE,不能不说是一个遗憾。

BERTScope在国内的用户并不多,而且仍然有亟待完善的地方。BUJ的注入解决啦,但分离现实中的BUJ并定位根源,仍然是一个问题。据说Agilent的86100D 可以进行BUJ的分离,他们8月份就有研讨会讲这个,期待。
附件: 您需要登录才可以下载或查看附件。没有帐号?敬请注册
2

评分人数

  • Triton

  • 袁荣盛

也就是说,QuickEye在几个小时内能仿真1e12的码源,不知道怎么实现,传统的spice不会有这么快。
如果人生也能仿真...

TOP

1e-300是啥前瞻?纯嚼头。以前系统的ber要求是1e-12,现在随着10G+互连时代的来临,要求达到1e-15。

不要认为QuickEye就很美好,原理上跟ADS Channel Simulation的Bit by Bit仿真模式一样,基于系统的LTI假设进行时域卷积运算,所以速度快。VerifyEye就是StatEye。

TOP

1e-300是啥前瞻?纯嚼头。以前系统的ber要求是1e-12,现在随着10G+互连时代的来临,要求达到1e-15。

不要 ...
beta 发表于 2010-7-20 21:52


反正,我看到这个指标我就傻眼啦。我们连1E-12都没有很好解决啦,BERT测1E-15都要3天半(10Gbps的速率)。人家就已经到1E-21啦,牛,咱只有望洋兴叹的份。这个差距比中美的差距可大得多,哈哈。

TOP

QuickEye Analysis

QuickEye analysis employs pattern-dependent convolution. A step response to a single transition is taken for a single rising edge and a single falling edge, and the step responses are then combined to create an intermediate waveform that approximates the behavior of the fully-simulated channel. With QuickEye, inter-symbol interference (ISI) is easy to calculate, but random transmit jitter is more difficult to handle. QuickEye superimposes the step responses for a rising and a falling edge in their appropriate places. The step response is the waveform at the far end of the channel to a single transition at the input, from low to high or high to low.

By default, Nexxim simulates over a time period of 100 UIs to calculate the step response. STEP_RESP_NUM_UI parameter on the Eye Source controls this operation.

QuickEye convolves the calculated step response with an input bit pattern to generate the eye diagram. The input bit pattern can be a literal bitlist specified in the netlist, a sequence of random bits, a bit pattern read from a file, a pseudorandom selection of bit patterns of length 7, 11, 23, or 31, or the worst-case bit pattern calculated by peak distortion analysis.



VerifEye Analysis

VerifEye analysis is a statistical eye-analysis algorithm, similar to the public domain StatEye .VerifEye calculates a cumulative distribution function (CDF) for deviations from the step responses in the intermediate waveform described above, taking into account the conditional probabilities for the various kinds of transitions. A slice through the CDF generates the bathtub curve. Moreover, random transmit jitter can be simulated by using a Gaussian distribution for the deviations in response. VerifEye analysis uses the edges for calculations (rather than the pulses as in standard StatEye). Using the edges allows random jitter to be truly random (timings of edges are independent) and also enables an easy calculation of duty cycle distortion (DCD).

The idea behind VerifEye is that in order to generate an eye diagram and bathtub curve for a channel, including the effects of jitter, it is not necessary to run a very long transient simulation, such as would be implied by the required BER of the channel. By combining the statistics of the bit stream with the variation in transitions due to jitter, it is possible to generate the information needed in much less computation time.

TOP

返回列表