Time to Get Ready for PAM4!

樱桃海弥 · 发表于 2016-4-21 16:55

EDA365欢迎您！

您需要登录才可以下载或查看，没有帐号？注册

x

本帖最后由 alexwang 于 2018-7-3 09:35 编辑

Time to Get Ready for PAM4!

In the course of our careers, we don’t get many chances to be trail blazers. With PAM4 design and test techniques still in flux and different labs using different techniques, it’s the perfect time to become a PAM4 expert.

在我们的职业生涯中，我们可能没有很多机会去当拓荒者。PAM4的设计和测试技术的正在积极的发展当中，是时候去了解PAM4了
Tektronix has just released the first, complete, up to the second, PAM4 application note: “[url=]PAM4 Signaling in High Speed Serial Technology: Test, Analysis, and Debug[/url].” It explains what PAM4 is, the problems it solves, the problems it introduces and what to expect going forward. They even promise to keep it up to date.

泰克刚刚发布了第一个关于PAM4的应用文档“PAM4 Signaling in High Speed Serial Technology: Test,Analysis, and Debug.” 在这个文档里，解释了PAM4是什么，带来了什么问题以及将来会出现什么问题。

Standards Today

So far, just one PAM4 standard has been released: 100GBASE-KP4, 100 Gigabit Ethernet, which is covered in IEEE 802.3bj. At 13.6 GBd (i.e., 27.2 Gb/s), it hasn’t been widely adopted because good old NRZ works okay at these rates. But there’s no getting around PAM4 above 50 Gb/s. The new PAM4 App Note comes from the experience of the Tektronix engineers who are working on PAM4 technology and contributing to 400G standards like 400 Gigabit Ethernet, 56G OIF-CEI, and all the rest.

现有标准

目前为止，只有一个PAM4的相关的标准发布，那就是100GBASE-KP4, 100Gigabit Ethernet。这个定义在IEEE 802.3bj，速率是13.6 GBd (i.e., 27.2Gb/s)。目前这个接口还没有被广泛使用，主要原因是NRZ在这个速率是还是可以工作。但是如果速率到了50 Gb/s以上，那就不得不使用PAM4了。这个应用文档来自泰克工程师，他们正在研究PAM4技术以及参与400G以太网标准，56GOIF-CEI和其他标准

How did we get here?

The problem is the frequency response of transmission channels. In the good old days, we could think of digital signals as though they were DC signals that turned on and off with logic-highs and logic-lows. But data rates increased and we had to think of high speed signals as microwaves propagating along waveguide traces on printed circuit boards. Conducting traces on circuits and backplanes are all twists and turns, skin effect and dispersion with loss of 70+ dB around 25 GHz. The resulting ISI (inter-symbol interference) shuts eye diagrams.

We got away with baseband-looking NRZ by conditioning the signal at both the transmitter with FFE (feed forward equalization) and at the receiver with CTLE (continuous time linear equalization) and DFE (decision feedback equalization) but these techniques run out of steam somewhere between 25 and 50 Gb/s.

我们为什么要用PAM4

问题来自于传输通道的频率响应。在过去的低速率的美好日子里，我们可以把数字信号想象成像直流信号一样，逻辑高时开，逻辑低时关。但是随着速率的增加，我们不得不把数字信号想象成一个微波信号随着波导走线在PCB上传输。在电路和背板上的导体走线，由于趋肤效应和介电损耗在25GHZ可以到达高达70dB的衰减。由此引起的ISI（码间干扰）会使信号眼图闭合。

解决这个问题，在像基带信号一样的NRZ信号上，我们发送端采用FFE（前馈均衡），接收端采用CTLE(连续时间线性均衡器)和DFE（判决反馈均衡器）来均衡信号。但是这些技术在25到50 Gb/s面临很多挑战。

Comparing PAM4 and PAM2-NRZ

By coding two bits per symbol, PAM4 transmits twice as much data as PAM2-NRZ at the same symbol rate, the same baud. You’ll notice that I put a “PAM2” prefix on good old NRZ. The thing is, the digital-looking NRZ signals are more accurately described as 2-level pulse amplitude modulation, PAM2, than non-return to zero.

PAM4和PAM2-NRZ比较

通过把两个比特变成一个码元，PAM4可以在同样的码元速率上传递两倍于PAM2-NRZ的数据。你可能会注意到我再NRZ前面加了PAM2。原因是数字的NRZ信号，用2电平脉冲幅度调制（PAM2）来描述，比非归零来的更准确。

The figures below compare PAM4 and PAM2-NRZ signaling.

下图是PAM4和PAM2-NRZ比较

Comparison of PAM4 and PAM2

PAM4 has sixteen different bit transitions compared to four, six rising/falling edges compared to two, and three eyes enclosed in the voltage swing and unit interval where we used to have one. PAM4 suffers at least three times the signal-to-noise problems that PAM2-NRZ faces. One symbol error can cause two bit errors, especially if jitter is the culprit. Along with these complications, we now have to think about the weakest link in the three eye chain; the BER (bit error ratio) will only be as good as the worst eye.

PAM4和PAM2-NRZ比较

PAM4有16种不同的比特翻转，PAM2只有4种。PAM4有6种上升下降沿，但是PAM2只有2种。PAM4在1个UI上有电压上3个眼，但是PAM2只有一个，所以PAM4会面临至少3倍于PAM2-NRZ的信噪比问题。一个码元错误可以会造成2个比特错误，特别是如果抖动是个问题的话。PAM4我们需要关注的是这三个眼中间最差的那个，因为这个决定了系统的误码率。

We will bring every tool that we have to the PAM4 lab bench. Differential signaling, embedded clocking and clock recovery, and equalization at both the transmitter and receiver. Differential signaling doesn’t change, clock recovery is trickier with edges that aren’t as distinct, transmitter equalization—both multi-tap FFE and simple de-emphasis—is complicated by the four symbol levels, CTLE receiver equalization doesn’t change, but DFEs have four decisions to feed back.

我们会把我们所有的工具带到PAM4的实验室。差分信号，嵌入的时钟数据恢复，发送和接收均衡。差分信号还是一样的，时钟恢复变得更加复杂，因为信号的沿翻转不是那么明显了。发送均衡，包括数字多拍的FFE和简单的去加重技术，由于是4个电平而变得跟复杂了。CTLE接收均衡没有变化，但是DFE有4个判决结果要去反馈了。

Three eyes also means three voltage slicers. The first adopters will use common timing for the three slicers; later technologies might have independent timing. The relative proportions of the three eyes introduces a whole new category of linearity issues. Right now, engineers in different labs use their own techniques to measure timing and voltage nonlinearities. As the technology evolves new test techniques will converge.

3个眼意味着3个电压判决器。刚开始大家会使用共同的时序的方法给这三个判决器，后来的技术可能会使用独立的时序。这三个眼的相对比例会引入一类新的线性度问题。目前，不同实验室的工程师们用他们的方法去测量时序是电压上的分线性度。随着技术的演进，这些方法最终会统一。

PAM4 might seem to create as many problems as it solves, but the standards bodies have granted us a boon: by incorporating FEC (forward error correction) the BER requirement has been relaxed by a factor of 100,000. Instead of designing and testing to BER better than 10-12 or 10-15, PAM4 designs have to meet BER < 10-6. Relaxing the PHY layer BER requirement gives one gargantuan test advantage, and we’ll need it.

PAM4看上去带来的问题，跟它所解决的问题一样多。但是标准组织都在引入FEC(前向纠错技术)，这样BER的要求可以放宽到100,000倍。PAM4的设计只要满足BER < 10-6，而不是10-12 或者10-15 。放宽了PHY层的BER要求给我们测试很大的便利。

To test PAM2-NRZ systems down to BERs of 10-12 or lower, we either had to perform tests that took 20 minutes to an hour or extrapolate faster measurements down deep into the noise. Where interpolation is a perfectly reasonable technique for estimating a value between two measurements, extrapolation is a leap off the edge of a measurement’s reliability. By requiring BER < 10-6, we can measure the full impact of noise and jitter in a fraction of a minute with a BERT (bit error ratio testers) or an oscilloscope.

要测试PAM2-NRZ系统的误码到10-12 或者更低，我们要么去做20分钟到1个小时的测试，或者推演的到噪声的方式快速测量。通过要求BER < 10-6 ，我们可以使用BERT（误码测试仪）或者示波器在不用到1分钟的时间来可以完全直接测量噪声和抖动的影响。