openwifi · 1月2日

一场关于开源硬件和软件无线电的讨(论)论(战)合集

2020年末发生在twitter的一场关于开源硬件和软件无线电的讨(论)论(战)。从我29号晚上8点半发出的一个twitter开始,到2021年1月1日凌晨ADI公司的人差点骂人结束。事情从ADI负责社区和开发板的Robin Getz火药味的回复开始,一场众多硬件maker和ADI这种巨型公司关于芯片价格、开源模式的唇枪舌剑。这里试图按照时间线记录一下,以提供不同的人的不同的立场和视角,供大家思考。关于开源软件、硬件、软件无线电和个人创客。

主要出场人物介绍:

我:一个只是想求证一下ADI官方hack芯片的方法如何理解的硬件小白

Robin Getz:ADI系统工程总监,负责参考设计,开发板,开源社区这块

Philip:FOSDEM软件无线电开发者房间组织者,GNUradio的人

Scott Torborg:国外硬件众筹网站crowdsupply的联合创始人,开源硬件创客圈内人

Alexandru Csete:应该是最早发现ADI芯片hack的人

F5OEO:rpitx作者,把树莓派IO变成FM广播发射机。喜欢hack各种硬件

OSS FPGA & ASIC EDA tools:开源FPGA/ASIC设计工具大V

Ben Hilburn:他曾经在DARPA(国防部高级计划局)组织频谱挑战比赛darpa spectrum collaboration challenge 2(用了非常多的USRP和射频电缆,构建云端的真实信号模拟环境)。三年的比赛结束后他去了微软云计算。我怀疑他要把这个真实的无线电仿真业务带到微软云上?

Derek Kozel:也是FOSDEM软件无线电开发者房间的主持人,GNUradio的人,硬件自由工作者

热心观众:sylvain azarian,Aaron Holtzman,Colby Boyer

因为近期在考虑设计openwifi FPGA dongle,最终发现我需要的只是一个7z020版的plutosdr。plutosdr是ADI公司自己设计的开发板。主芯片是xilinx 7z010 fpga 和 ADI公司的ad9363。而我们的openwifi设计需要规模大一点的7z020 fpga。ad9363芯片很廉价,但能hack成ad9364(adi的wiki有官方教程),hack之后频率就能从之前的3.8GHz扩展到6GHz。我的疑惑是ad9363是双通道,ad9364是单通道。hack了之后,难道禁用了一条通道?

于是发了一条twitter表达我对于通道数量的疑惑,并且at了发现这个技巧的Alexandru Csete,FOSDEM软件无线电开发者房间的组织者Philip(对硬件很熟),以及ADI负责社区以及开发板的Robin Getz。

很快热心观众sylvain azarian帮我ping了F5OEOEvariste,F5OEOEvariste说ad9363实际上是可以hack成双通道6GHz的ad9361的,改driver即可。我向他表达感谢。

事情到这里一切正常,直到ADI公司的正主Robin Getz回复:

这不是hack,这是又多了一个对软件撒谎的人类(软件本来是信任人类的)。

(It's not a "hack". It's just one more human lying to software that trusts you.)

Robin Getz接着说道:
AD9361/3/4寄存器完全一致。驱动读取芯片的产品ID(0x037),但三种芯片返回的ID是一样的。驱动知道不同芯片的唯一途径是编程的人通过设备树告诉它
(The AD936[134] all share a common register set. The driver reads the "PRODUCT ID" (0x037) buts it's the same on all parts, the only way the driver knows the difference is a human telling it, via device tree)

Robin Getz接着说道:
这就是Pluto启动时的魔术(magic):可以在内核启动之前方便的改动device tree。我们在嵌入式Linux会上的演讲对此有详细解释:
(That's all the Pluto boot magic does - is have a convenient way to patch a device tree before the kernel boots. We detailed it in this presentation at the embedded linux conference:)

Robin Getz也许意识到了我是想自己做廉价硬件,他接着“警告”:
对软件撒谎长期来看会有问题,因为没有保证、质保。也许你第一次很幸运,一切都很正常。但下一次也许就不那么幸运,不工作了。也无法退货保修什么的。
(lying to software - will cause problems long term, there is no guarantees/warrantees on it. You may get lucky on first build & everything is OK, then on the next build - be not lucky and have all of them not function out of spec, and you can't get a refund/repair/different batch.)

F5OEOEvariste出来打圆场:
通过对软件撒谎来做一些有趣的事情,我意识到可能最终会失效。比如rpitx(能把树莓派变成FM发射机)也是通过不按数据手册来实现的。不按数据手册来,才使得我们发现了plutosdr原来而已变得这么强大(意译)
(If fun things could be done by lying to software, I am aware that it could eventually fail. rpitx is also made mainly "out of spec" : just let us discover the fabulous limits we have with plutosdr platform.)

Robin Getz进一步点我的名:
同意!做有趣的事情和做产品是两码事。他jxjputaoshu是想做产品。做产品需要很多努力。从一开始就想着不按手册从来不是一种开发产品的好方法。
(agree! Having fun and making products are 2 different things. jxjputaoshu
is asking since he wants to make a product; Designing things for production takes a lot of effort, and running parts out of spec at the start isn't ever a good way to develop things.)

我说:我需要的仅仅是一个7z020版的plutosdr。

Philip出来了,他应该是知道Robin总是很愤怒,他发推调侃:

某些人从来都是不高兴。

( Some people are never happy :) )

接上条

我对Robin表示礼貌性感谢,继续解释道:

现在我明白了为什么我们openwifi的设备树里全是写的ad9361,但是也能在ad9364芯片上工作。

Robin Getz继续“政治正确”:

你这么做可能会损坏设备。不同芯片的PCB设计是不同的:比如AD9364的一些管脚是接电源和地的,相应的功能也没有在驱动里开启。如果你开启了,但PCB设计并不支持,这就没有质保了。如果你需要AD9361,去买AD9361.

(doing so could damage the device. PCBs are designed differently - and AD9364 has pins tied to VDD/GND; and their functions are not powered on in the driver. Turning something on - when the PCB is not designed that way - voids the warrantee. If you need a AD9361 - go buy a AD9361.)

Robin Getz继续抓着我说的“我需要的仅仅是7z020版的pluto”,他说:

那也就是Ettus E310,或者ADI ADRV9364-Z7020模块。很快你可能会说你不想为额外的温度范围、内存、FPGA资源、ARM核心、电源、射频通道、硬件开发和验证成本付钱。

(That’s an Ettus E310; or an ADI ADRV9364-Z7020 RF SoM. Soon you will just say that you aren’t willing to pay for the extra temp range/memory/FPGA/core/power/RF/verification/engineering that goes into those...)

他说的这两个东西一个三千多美金一个一千多美金,真是大公司待惯了,以为大家的钱都是大风刮来的吗。

对话进行到这里,我明显感觉到Robin Getz不知道哪根敏感的神经被触动,进入对我的嘲讽模式。可那个Hack是ADI官方wiki给出的,他不应该如此“愤怒”。现在回想起来,他可能“愤怒”于,又有一个家伙来山寨我们了,还好意思向我们请教!

Philip看不下去了,说:

我觉得那正是他(我)说的。他的目的就是使得人们能够以Plutosdr的价格,更容易的hack一个完整的开源WiFi实现。

I think that is exactly what he is saying :) His goal is make it easy for people to hack on a complete open source wifi solution for about the cost of a pluto.

Philip很显然是明白openwifi项目的意义的。去年年初布鲁塞尔FOSDEM时我就是在他组织的软件无线电开发者房间官宣了openwifi,并给了半小时报告。而且他很清楚这一年来想玩openwifi的人很多,但动辄1000美金以上的FPGA开发板,不是人人都能有预算。plutosdr只要一百多美金,可惜FPGA 7z010太小,装不下我们的设计。所以我才说“我需要的仅仅是一个7z020版的plutosdr”

未完待续。

接上条。

因为我们主要搞的是所谓软件无线电,我说:

软件往往意味着无休止的补丁和升级。不像硬件,发售后就无法更改。

Robin Getz显然对迭代式软件开发不屑:

听起来像Windows程序开发者。嵌入式开发者不这样。软件并不总意味着补丁和升级。软件的确容易做新特性,但这也是为什么任何项目开始时spec锁定是很重要的。

(Sounds like a Windows application developer - not an embedded developer. "software" does not not always imply patches and updates; new features are easier to do in software; which is why specification lock down is so important at the beginning of any project.)

毫无疑问这是芯片和硬件的哲学。ADI的做事习惯。

我再次重申,openwifi为什么需要7z020版的plutosdr:

因为其他昂贵的开发板上有很多openwifi不需要的功能,比如音频、视频、hdmi、光纤、pcie、usb等,plutosdr没有什么多余的外设,很便宜,很适合openwifi,但很可惜7z010的fpga太小。

Philip其实看到了事情的本质,他说:

硬件和开发都是需要钱的。openwifi dongle的价格目标很大胆(接近plutosdr),但我怀疑plutosdr之所以能那么便宜,是因为ADI的开发时间没有算在内。

Hardware and engineering cost money. You have set a bold goal, I suspect Pluto's price (point was possible due to ADI basically donating engineering time.)

Philip说的很有道理,本质上ADI做plutosdr就是顺手的事情,芯片又是自产,当然能定价很低仍然有利润。又当裁判员又当运动员的意思。

听了上面的话,Robin Getz貌似理解为别人为ADI感到自豪?被触发了炫耀模式,他说:

你也许会震惊:ADI的10年毛利率大约是60.39~68.28% (中位数65.44%)。在817个半导体公司里属于top 6%。如果没有可靠地商业模式是不可能取得这样的成就。

You might be shocked[1] - ADI's Gross Margin over 10 years has varied between 60.39 and 68.28% (Median: 65.44%). This is in the top 6% of all the 817 companies

in the Semiconductors industry. You don't achieve that by not having a solid business plan.

这也是ADI建立ADI基金的原因。是通过有意义的方式扩展公司的社会影响力。绝不仅仅是给某个人捐赠几个芯片,而这个人完全不会帮助解决全球性或者系统性的问题。

Which is also one of the reasons ADI started the ADI Foundation. To expand the companies social impact in a meaningful way (more than a few part donations to specific people that doesn't really help global or systemic issues).

Getz一不小心又开启晒优越,对我们这种小人物的嘲讽模式了。我心说你咋不去竞选总统呢。你ADI公司很伟大,芯片很伟大,还尝试解决全球系统性问题?我们做的openwifi就是nothing?

说到这里,各位可能会有点懵,说实话我也有点懵。我觉得这位Getz先生更懵。openwifi如果发展的好,还不是ADI可以卖更多的AD936x?对用了他们芯片的项目火力全开极尽嘲讽,这可真是傲娇的很那。

接下来,Scott Torborg加入讨论,这位看起来像是crowdsupply的co-founder。crowdsupply是国外最大的硬件创客的众筹网站,对于小规模硬件,个人硬件领域那是很熟悉了。他其实进一步说明了事情的本质,开启声讨模式:

这不是一个公平竞争领域:ADI以99美金发售PlutoSDR,这使得其他所有人再做基于AD936x的商业可行的硬件变得很困难。就算假设最高采购量,从公开零售渠道买plutosdr的主芯片(7z010+ad9363)已经137.75美金了。

It's also not a level playing field: ADI launching the PlutoSDR at $99 made it really hard for anyone else to make a commercially viable, inexpensive AD936x radio. Even the public retail pricing at the highest qty tier for XC7Z010-1CLG225C and AD9363 *alone* is $137.75.

事情到这里,风向有点变了。看来大家对于ADI的傲慢积怨已久。

未完待续。

接上条。

Scott Torborg(crowdsupply co-founder)作为创客这边的代表,继续提供建设性讨论,他说:

如果把射频板FMCOMMS2(也可以用ad9363,一般用也够了)降价到plutosdr以下,然后人们就可以用手中的zedboard来玩openwifi了。

It would be nice if even just the FMCOMMS2 (which could use an AD9363 and still make everyone happy) were priced at or below the Pluto. Then people could use their Zedboards and such for OpenWiFi.

这的确是个办法。zedboard是纯fpga板,用的是7z020 FPGA,容量是可以跑openwifi的。zedboard也是低端产品,比较便宜,配上FMCOMMS2这种射频板就能用。但ADI的FMCOMMS2射频板还是挺贵的,比plutosdr(fpga+射频)还贵不少!虽然射频板上用的芯片和plutosdr里的射频芯片hack之后相同。

其实Scott Torborg揭开了ADI的遮羞布。从上面荒唐的定价策略(单独射频板比FPGA+射频还贵不少)就能看到出来,plutosdr就是用来屠杀一般创客,阻止个人来基于ADI的芯片build自己的个人硬件的。

Robin Getz当然有冠冕堂皇的理由辩解:

我们在美国生产了几百块fmcomms射频板(为了管理供应链问题),而且用350K美金的设备做了严格的射频测试。测试全是手工的(因为量不足以上自动化)。硬件不只是BOM。

We make fmcomms boards by the hundreds, in the US (to help manage supply chain issues) and do much more rigorous RF testing (connected to a $350k of test equipment). Test is all manual (it’s not high enough volume to automate). There is more to things than BOM.

他说的的确是在理的。就像每次Apple新机发布,都有人分析BOM成本和Apple的暴利。但他言语之中就是plutosdr没有这些成本,所以便宜?个人觉得这些有一定道理,但更多的是说辞。ADI的射频芯片集成度非常高,硬件按照参考设计来翻车的可能性很小,芯片外部元器件很少。没必要每个板子都做昂贵的手工测试。

所以想贵就贵,想便宜就便宜。谁让人家是芯片原厂,有掌控权呢?

未完待续。

接上条。

至此开启更多芯片定价行规内幕讨论。

Scott Torborg问robin getz:

好奇,有一款芯片叫AD80403是不是对于他(我)的WiFi项目更便宜?

just curious, do you think the AD80403 could be a possible / more economical option for the WiFi project that jxjputaoshu is pursuing?

这款AD80403芯片如果你搜索的话几乎无任何资料,只有两个地方出现:1 ADI的无人机通信系统方案;2 知乎上对大疆无人机图传硬件拆解。看功能和指标和AD936x很像,但是阉割了大部分频率,只留下了WiFi的2.4GHz和5GHz。

所以Getz也蒙圈了,他说:

我也仅仅能找到ADI官方无人机方案手册,我从没听说过。也许是半定制的,不公开的。

I only found info on that in ... I haven't actually heard of it before. It might be a semi-custom, that isn't available widely.

和几个朋友讨论下来,感觉这可能是大疆定制廉价版,也许只是一个代号而已,区别于ad936x,这样防止被抄袭?总之神奇的事情可以在中国发生,让老美蒙圈去吧。

关于前面Scott Torborg提到pluto的bom成本比售价还高,我问我发现最近fpga价格大涨,是不是因为比特币又上来了。

Robin Getz立刻又拿他的业内专业性来教我做人,他说:

这纯粹是市场正常波动(供应/需求/提前期)与比特币开采无关:)还给了一个关于大规模芯片短缺的新闻报道。

This is the pure commodity driven / interrelated supply / demand / leadtime / pricing curves sliding up and down (and doesn't have anything to do with bitcoin mining):

Scott Torborg再次紧咬价格不放,他说:

Xilinx FPGA价格在美国远高于其他地方。但就算我们拿到很优惠的AD9363价格(tray price),他依然达到80%的pluto售价,和pluto竞争太难了。

Yeah, Xilinx FPGA pricing is all over the map, and certainly far higher at US retailers than anywhere else. But even the current tray price for an AD9363 is still 80% of the Pluto's launch price. Very hard to compete with.

此时另一个大V:OSS FPGA & ASIC EDA tools 加入讨论,他说:

唯一知道项目是否可行的办法是从ADI拿到一个针对该项目的价格。

The only way to make such a design project feasible is to get a project price from AD for the chip.

我觉得有点废话的意思。应该是年末都在休假,大家都有闲工夫在twitter打嘴炮。

Getz接着废话:

和本地销售签订NDA,拿到批量价格。去亚洲找到低成本制造方案。开发一套低成本测试方案(批量测试)。找到仓库存放10k 片板子,并接受每月的库存成本。

Contact your local sales to get an NDA and volume pricing. Then travel to Asia to work out low cost manufacturing. Then develop a low cost high volume test solution. Then have the warehouse space to put 10k units, and the fortitude to eat monthly inventory costs.

我敢说大部分创客一开始肯定不是这么干的,而是打板,然后手工焊接调试。当然Getz先生干一定是高大上的干法。

Scott Torborg再次把大家拉回来,他说:

我不是反对规模经济或者ADI在plutosdr上的巨大投入。只是觉得这样的后果是:开源硬件项目很难拿到批发价。

I certainly don't deny the economies of scale or the large investment that ADI has put into PlutoSDR, to great effect: I'm just noting that it will be very difficult for an open hardware project to match the "volume pricing" part of that.

我说某些项目还是达到了大规模,比如raspberryPI。

再次被Getz鄙视,他说:

说实话,raspberryPI和其他大多数小型硬件项目是不一样的。一年卖出6百万个PI。对不起,我不认为openwifi也能这么火。

Let's be real - Raspberry Pi is in a different category than most products; Comparing it to other small hardware projects isn't good in any way. Pi sales are about 6 million units per year. Sorry - but I don't think there are that many people interested in opensource Wifi.

这个太抬举我们了,现在阶段我们达到PI的百分之一,千分之一就是不敢想象了。

接着Getz回应对于Scott Torborg关于价格的追问,他的回应还是老一套:

BOM不是项目成本的全部。对于小批量太多其他因素可以主导成本了。

The BOM is only a part of the overall cost of a project; and in low volume ones - it can be everything else that dominates the cost, or sets the price of a product.

Getz先生仍旧是无比正确!

关于ADI的plutosdr价格,OSS FPGA & ASIC EDA tools继续深挖,他说:

通常售价至少是4倍的BOM成本。所以PlutoSDR BOM大约25刀。(因为售价100刀)

Usually the sales price should be at least 4 times the BoM cost. With this thumb of rule, the BoM cost of Pluto SDR is around 25 USD.

Getz不干了(我们可没这么便宜),他说:

不知道你哪里听说pluto价格100美金,我们的官方售价249美金。只是三年前对于初始用户有优惠是99美金。

The commercial price (from ADI) of Pluto is $249. I don't know where you are getting $99 from (yes - that was the introductory price 3 years ago; but that was just taking a page from the crowd-funding web sites; offering a deal to initial users).

一些分销商(mouser和digikey)为了获取教育领域的新用户,学生折扣后卖150美金,这是他们的商业模式。

And yes - some distributors (Mouser and Digikey) who have an academic/learning focus - offer student discounts; and it's pretty common to find it at $150; but that's part of their business model, (new customer acquisition is worth $ to some). It's just less margin, not less cost.

Philip继续感叹软硬件模式的不同,他说:

一个智慧的人曾说,开源软件行的通是因为可以零成本复制软件。制造硬件却需要真金白银,事情会变得复杂。

A wise man once told me open source software works because the cost of making copies is virtually zero. Making hardware costs real money and this makes things complicated.

Getz再次以他们的“硬”文化自豪,他说:

早发布,勤发布,让用户测试 -- 这些对硬件都不成立

release early, release often, let your users be your testers - doesn't work for hardware either.

这时,一位无奈躺枪的“受害者”出来了,Charles Brain,之前搞了自己的一套SDR硬件,以及数字电视调制解调软件,结果Limesdr和Plutosdr出来之后,卖不出去了,因为这两个价格太低了。他说:

我们的DATV-Express无法在价格上和Limesdr、plusdr竞争,所以现在不做了。我们仅仅是找到了一个市场的缝隙,然后去做。提示:如果你的产品能填补一个市场缝隙,那么价格就不那么重要。

Interesting thread, we ceased producing DATV-Express boards when Lime/Pluto were released because we couldn't compete on price. As we were only trying to fill a gap in the market we were happy to do so. Tip, find a product that fills a gap and the price becomes less relevant.

其实我不太理解他说的,他应该感到很无奈。他的数字电视产品显然还是填补了市场的空隙,结果因为大公司的低价强势进入,无辜躺枪退出。

未完待续。

接上条。

Philip 看到那位无辜躺枪的感慨,同感慨:

找到市场缝隙并填上,理解硬件定价游戏。请记住第一代PI基本上是报废芯片做的。每一代设计也有严重的设计上的妥协。

Find gaps and plug them, understand the hardware pricing game. Remember the first PI was (basically) made from scrapped chips. Even later models have some serious design compromises.

Scott Torborg继续cue ADI,说道:

这里硬件定价并不神秘。我只是抱怨这种模式对于小批量开放硬件的项目并不容易。如果ADI可以低价销售大量报废芯片,那就太好了!

Appreciate the sentiment, but the hardware pricing game isn't a mystery here. I'm just complaining about it not being conducive to people trying to support their open hardware dev efforts with low volume sales. ADI selling a bunch of scrapped transceivers would be great!

Scott Torborg继续,说道:

我希望更多芯片制造商能够像博通那样支持开放参考设计,社区项目。开源项目贡献者无疑正在做很多事情回馈他们(芯片制造商)。

Yeah, my hope is that more chip makers follow in Broadcom's footsteps in supporting open reference designs, community-oriented projects, etc. Open source contributors are certainly doing a lot of work that benefits them (specific chip makers).

Getz又忍不住秀ADI的社会贡献,说道:

ADI对开源的贡献很大,而且也有开放参考设计。从Linux驱动到Arduino配件,PI的配件,Jetson板子的配件,免费的仿真软件LTSpice。ADI作为外围/周边供应商,很难成为一个平台(这句话也许翻译不准确。我理解大意上ADI不是CPU/处理器厂家,所以自己很难成为平台,也就是作为硬件的中心。ADI只是周边,比如射频芯片)

ADI make lots of investments in open source; and open reference designs; From Linux drivers to Arduino Shields, Pi HATs, Jetson addons, (with SW) to zero cost LTSpice. It's just hard to be the "platform" as the peripheral vendor.

Scott Torborg顺势捧ADI,说道:

当然,我并没有排除ADI的意思,或者无视ADI的贡献。我从ADI的设计资源里学到很多,从你的书,从Jim William的应用笔记。最终目的,我只是想看到更多的人可以打造更多有趣的无线电硬件(如果是开放license就更好)

Absolutely, I wasn't trying to call ADI out, or diminish that investment. I've learned a ton from resources provided by ADI, from your book to Jim Williams' app notes. Ultimately, I just want to see more people build interesting (and ideally, open-licensed) radio hardware...

Scott Torborg接着点明了一个异常残酷的事实:

现在开源硬件/FPGA项目,如果不卖硬件那么往往难以为继。卖硬件之外的其他盈利模式并不十分有效。(他再次cue回ADI的AD936x)我已经听好多人说他们希望制造相对不贵的基于AD936x的硬件。

At the moment, it's hard to sustain work on open source hardware/FPGA development without selling hardware: the alternate revenue models just aren't very effective yet. I've heard a number of people say that they want to market a relatively inexpensive thing based on an AD936x...

不得不说Scott Torborg作为硬件众筹网站创始人,对其中的生态是有一定洞察的。的确如此,对于硬件项目,往往需要靠卖硬件才能不亏和生存。这里卖硬件之外的其他盈利模式,应该是指捐赠,或者类似于patreon这样的网站的模式。

我表示同意。很多开源项目的维系依赖硬件销售。很多人被迫做硬件。比如openwifi,如果市面上有便宜的硬件可以让大家毫无压力的购买,我们绝不会碰硬件。硬件显然不是我们的强项。我们目标在于发布能跑在各种FPGA开发板(包括最廉价入门级)上的wifi参考设计,创建社区和生态,鼓励更多的人基于我们的参考设计,打造各种有趣的开放的系统甚至芯片。

Scott Torborg再次确认,说道:

完全同意。这是另外一件事情我希望本来可以不一样的。捐赠,惠顾,支持,项目资金,仍旧很困难。(如果开源项目不用做硬件就能靠这些来源持续,那将会很好。大家可以focus在自己擅长的事情上,而不是被迫去做硬件卖硬件)

Yeah, totally. That's another thing I wish was different. Donations, patronage, sponsorship, grants are all still really hard to make work.

Getz当然也很清楚这种软硬件项目的不同,他说道:

企业用开源软件的盈利模式已经找到了,而且有时候是可行的(我理解,比如redhat,还有那个开源数据库什么的)。对于嵌入式软件的开源盈利模式仍旧很模糊。最终消费者需要的是消费者模型:拿钱换物,之后是永久的免费支持和维护。

yes - the economic model of open source software (enterprise) has been figured out and works (sometimes). For embedded - I think it's a little more murky. End consumers want a consumer model - exchange of $ for a widget, and expect future/forever maintenance & support for free.

我想这就是2B和2C最大的区别。目前开源软件盈利很好的可能都是2B的。

但我并不同意Getz说的嵌入式开源软件适用于消费者模型。我觉得SDR板子的用户并不是最终消费者,而是大量的开发者。开发者做二次开发,然后试图再次赚钱,或者仅仅是觉得有趣,爱好,自己玩,自己用。就算不做二次开发,大多数买来SDR板子的,也都是要试试其他人二次开发好的软件(比如ADS-B软件,gnuradio,FM收音机什么的)。比如openWRT的用户,一般是买来折腾的。或者有些公司折腾完了,包装一下再卖给最终消费者。最终消费者并不关心里面是否openWRT。

然后Getz开始隐式嘲讽(我玻璃心了也许):

关于开源商业模式James Bottomley有一段话很有趣:

  • 伟大的艺术家可以挣钱(毕加索),也可以贫困潦倒(梵高)
  • 成功的项目并不保证成功的商业模式

Interesting read from James Bottomley

  • it is possible for a great artist to make money (Picasso), but it’s equally possible to live their lives in penury (van Gough).
  • a successful project in no way guarantees a successful business model

Scott Torborg继续提醒大家注意Plutosdr的负面影响,说道:

但现在一切都被阻止了。因为plutosdr降低了大家对价格的期望,同时创客们拿到的小批量价格根本打造不出能跟plutosdr竞争的硬件。不要误解我,plutosdr本身很棒,价格也很低。

But that they've been deterred from doing so because the PlutoSDR has set price expectations bar so low, and they can't buy the chips in low volume at a price that lets them compete with that. Don't get me wrong, I think the PlutoSDR is awesome, and appreciate the low price...

Scott Torborg继续耿耿于怀,说道:

很可惜plutosdr带来了那样的负面影响。我想如果在开源软件无线电世界有更多选项,那么将会启发有趣的创意,并帮助人们更充分的利用收发机(射频芯片)。

It's just a shame that it has also had that effect. I think that more options in the open SDR world would really spark interesting ideas, and help people get the most out of the transceiver...

Scott Torborg继续开喷:

PlutoSDR, bladeRF micro, B200, E310, ADRV9361, Sidekiq, 这些硬件都很有趣,但他们用途不尽相同,方式也不同(这句话不太理解。大意是,都不完美吧,都有面向自己市场的特殊设计。有些太贵了)

PlutoSDR, bladeRF micro, B200, E310, ADRV9361, Sidekiq, etc: they're all interesting, but they all exercise different subsets of the Catalina capability, in different ways...

所以,当我说希望公司能够跟随博通的脚步时,我想说的是平价定价。这是开源硬件靠自己生存的所需要的。

So when I say that I hope companies will follow in Broadcom's footsteps, really what I mean is provide pricing parity, because that's what open projects seem to need in order to be sustainable...

我承认PI可能不是一个好例子,但我有一个强烈的印象,如果PI不能从博通得到便宜的芯片,这个项目就不会存在。

I agree the Pi is not necessarily the best comparison example: but I've gotten the strong impression that the Pi wouldn't have existed without an opportunity to get cheap Broadcom chips.

这里不仅仅是抱怨ADI一家。在芯片工业里,这样的问题(陡峭的价格曲线)广泛存在。Xilinx可能做的最差/狠。

I'm also not just whining about ADI here: the steep price curves everywhere in the semiconductor industry make this a challenging issue. If anything, Xilinx is probably the worst offender.

这时,估计是一位有深圳经验的人Aaron Holtzman 跳出来说:

在中国大陆打造硬件的公司,即使是小批量也能拿到很便宜的价格。这和西方制造的产品很不一样。

The silly thing is that companies that build their product at a CM in Asia don’t pay the list price even at low volume. So this a big deterrent to western made products.

中国,一片神奇的土地!

未完待续。

接上条。

Robin Getz终于开始回应Scott Torborg关于plutosdr价格屠夫的抱怨,但还是在打太极:

问题是做到便宜是很困难的。我理解你觉得小批量芯片价格过高导致亏损(或者少赚钱,或者不得不定价高),但实际上不是的。但我不能给你看BOM,因为我和其他人有NDA。这个行业里从来不会公布批量采购价!

The issue is the economics are hard. I understand it looks like this is a loss leader (it's not); but I can't show you the costed BOM without violating NDAs from everyone else. Published volume pricing doesn't exist in the industry.

好家伙,谁让人家是行业leader呢。上面这段话翻译过来就是:你说的不对,你不懂,我也不允许告诉你为什么!

Robin Getz继续补充:

这里面不存在孤岛。就在ADI发布Pluto(2017年3月)之后几个月,limesdr mini就发布了,早鸟价格99美金。(也曾定价75,83,现在175)

And, no one exists as an island; ADI introduced Pluto (Mar 2017); and a few months later - there are similar devices (at a glance) from other vendors that are in similar price points (early bird (Sept 2017) price was $99 (£75, €83), now $175)

上面这段话意思是,别老咬着我啊,还有另外一个家伙和我一样坏。

Scott继续从另一个角度旁敲侧击Getz:

我不是建议Pluto应该卖的贵一点。它对于教育很有趣,专业用户也能玩起来。我也相信Plutosdr对于ADI是赚钱的。

I don't mean to suggest the Pluto should be more expensive: I think it's enabled all sorts of interesting uses and education, and even "pro" users are often happy with the spartan features. I also trust you that it's been profitable for ADI.

这时候Getz是承认赚钱呢,还是只是交个朋友,断了大家的念想呢?Getz不再回应。

Scott继续总结强调他的观点:

我只是觉得有些可惜:这些小批量项目不得不在陡峭的价格曲线上去挣扎。这些独立的创客们,即使有众筹的帮助,也得不到同样的批量价格或者谈判价格。

I just think it's a bummer that new open or small volume projects are faced with competing across that pretty dramatic volume pricing curve. "Indie" project creators just don't have the same options to bootstrap volume buys or negotiate pricing, even with crowdfunding.

归根结底Scott仍旧在试图逼宫Getz。此人很会压迫人啊。牛。

这时候另外一个大V Ben Hilburn来了。他曾经在DARPA(国防部高级计划局)组织频谱挑战比赛darpa spectrum collaboration challenge 2(用了非常多的USRP和射频电缆,构建云端的真实信号模拟环境)。三年的比赛结束后他去了微软云计算。我怀疑他要把这个真实的无线电仿真业务带到微软云上?

Ben Hilburn说道:

一直在关注这个讨论,真的很有意思。我理解Scott的主要感叹是,要想制造pluto这样的低价硬件,你必须体量足够大。对吗,Scott?

Been tracking this since the thread got going. Really interesting discussion!

My read is that what Scott is really lamenting is simply that you must be "big" (many defs of that) to get the sort of pricing that makes an affordable PLUTO-like BOM possible.

Yes/No, Scott?

未完待续。

接上条。

Scott回应Ben Hilburn:

算是吧。我主要的担忧是,需要“足够大”会吓跑很多独立开源硬件开发者。

Sort of, but my concern is specifically that the requirement to start "big" discourages independent and open source development.

这时有深圳经验的Aaron Holtzman出来提醒大家:

如果你在中国,那么小批量也能价格很低。

另外一位网友Colby Boyer指出价格可能是因为第二供应商问题:

能产出和ADI类似射频前端的只有limemicro(就是英国limesdr那个),但这个公司比ADI公司小太多了。9361对于消费类产品太贵,对于高端应用太差。

I think part of it is there isn’t a real strong second supplier to ADI in this space. It’s Limemicro or ADI; two different classes of companies. The 9361 sits in a funny space: too expensive for consumer products and too poor performance for nose bleed applications.

Robin Getz给上面网友以出无情反驳,说道:

其实有很多类似AD9361的芯片。但人家都不像AD9361这样公开卖。人家(比如说)百万量(甚至更多)起订。觉得对于公开市场付出参考设计、软件、文档,不值当的。

There are lots of suppliers with products close to AD9361, it's just they are not open market (and require 1M units or more). Those vendors have determined that the market size is not worth the effort of putting together all the support SW/doc to handle an open market device.

好吧,大家应该懂得感恩ADI!(不过也是实情,AD9361参考设计、资料、代码都很不错。的确是给不在业内的大家开眼了,学习了)

Scott继续:

这是开源硬件世界(尤其西方国家,他知道深圳的生态好很多)需要解决的问题。

Something I'm getting at here is that this is a challenge that the open hardware world (especially in western countries) needs to solve.

Getz继续借力打力:

这是行业现状:BeagleBoard/TI; Raspberry Pi/Broadcom; Jetson/Nvidia; Cora/Xilinx; De10nano/Intel,以上是知名开源硬件。大家在这些硬件基础上做创新,而不是打造新硬件与他们竞争。因为人们如果考虑制造和购买,越来越多的人喜欢买。

Yes, it's something that spans across the industry BeagleBoard/TI; Raspberry Pi/Broadcom; Jetson/Nvidia; Cora/Xilinx; De10nano/Intel; Open hardware innovates on top of these platforms; not competes with them. More and more people look at make vs buy, and just want to buy hardware

废话。如果价格和功能合适,当然买。现在问题是你的plutosdr不合适跑我们的openwifi啊,我们被逼无奈去自己造。

Scott继续一口咬定ADI定价不地道:

你也许是对的,市面上有很多选项。但如果AD936x在美国能很便宜的买到,或者你的ADRV9364-Z7020模块价格是199美元,我一开始就不会有任何抱怨。

I mean, I think it's both: but you rightfully point out that there's a big spectrum of both integration levels and price curves. If the AD936x were cheaper at US retail distributors, or if the ADRV9364 were $199, I probably wouldn't have complained in the first place.

Scott完全理解openwifi的需求。就是7z020加射频。现状是很便宜的pluto是7z010加射频,而满足需求的ADRV9364-Z7020模块加载板售价超过1000美元(这个配置也就相当于把pluto里的7z010换成7z020),这也是为啥说你这个模块卖199美元,就皆大欢喜。

Scott继续指出基于Pluto打造新硬件很困难:

你的Pluto并不像你列出的其他平台,基于pluto打造可销售的硬件是困难的。(我理解Scott的意思是其他平台BOM和售价是合理的,但pluto的BOM和售价十分不合理,大家修改pluto设计打造自己的硬件,性价比完全和pluto不可比)。而且pluto集成度不高(我理解板子有点大,并不是一个小模块,不适合嵌入其他产品)

Unlike those other platforms you listed, it's a bit hard to build sellable hardware "on top of" the Pluto. It has 50 ohm coax (the ultimate standard) but like you were saying earlier: sometimes the optimal integration level is the single chip.

Getz继续顾左右,说道(Scott不是不满意pluto吗,我给你找找):

是的,和pluto类似的硬件是有的,比如Sidekiq Z2,这个是个小模块,适合做产品,不像pluto设计是为了学生扔进背包。

yes, those are more "modules" on the spectrum than pluto is. (for pluto's educational purposes, it was something that you can toss in a backpack, and come out the other end undamaged). The Sidekiq Z2 is more on the "module" side if that is what's needed

Getz这完全是深谙混元太极功法,接化发。他说的这个模块很昂贵!言外之意好吧这次BOM和售价合理了,Scott你满意了?

好吧,我承认Getz和Scott吵架不分上下,你们都很厉害。

然后Scott说出了我最想说的话:

最大的问题是,ADI,Xilinx这种公司到底希不希望出现很多基于你们芯片的开源硬件呢(你们给个人创客销售天价芯片)?而且想openwifi这样的项目目前只支持AD936x,这对于ADI是很大的好处啊。

One big question here is simply whether companies like ADI, Xilinx etc even want more open hardware products. But I'd argue that when something like openwifi only runs on AD936x, that's a pretty big feather in ADI's cap, and benefits them.

这也是我最想不明白的地方。你ADI就算瞧不上openwifi,也不至于这样冷嘲热讽啊。目前openwifi支持的6种平台,全是用的AD936x!对你ADI只有好处,没有坏处!

最后Getz再次打太极,说道:

开源硬件当然很好啦。如果你们开源界能搞定所有,这样我这边的工程工作就不用那么累啦。但实际上,我这边的工作是很底层的,一点都不有趣。我们得给制造的芯片写文档,我觉得你们都不会喜欢。

Yes, open hardware is great - It would be great to cut back on engineering & have it all done by "open source". But, the things I work on are low level, and are not fun at all. It's filling in the missing doc that doesn't come from the chip team. I don't think anyone would do it.

Getz最后开始显式嘲讽模式,说道:

好吧,来加入我的工程团队,(接下来就是说反话来讽刺浮夸的开源界),你会发现十分有趣:深入底层,反向工程缺失的不完整的或者不精确的数据手册和文档,我们还会把一些异常功能当做有趣,而且你会十分满意你是第一个使某种东西工作的人。

Oh, yeah - come join our engineering team - you will have lots of fun, working on low level pieces, and reverse engineering missing/incomplete, or inaccurate datasheets/doc. We put the fun in dysfunctional. It's satisfying knowing that you are the first to make something work.

我来稍微解释一下,前面一段Getz说你们开源界水平不行,搞不定我的工程工作(你们能搞定最好了,我就轻松了)。我的工程team很枯燥的。然后开启赤裸裸反话嘲讽模式。哦,对了,加入我的team,你可以“深入底层”,就是指开源或者黑客界爱号称自己深入硬件什么的。“反向工程”,这是指由于芯片的不开放开源界很多时候不得不去反向猜测芯片的接口和行为,然后才能写开源驱动。“把异常的功能当有趣”,我理解,比如把树莓派IO变成FM发射机,hack 9363成为9364,这些小把戏。“第一个使某种东西工作的人”,是的有时候开源或者黑客界喜欢制造耸人听闻的大新闻。

Getz这是最终通过赤裸裸的鄙视(距离谩骂不远了)开源界结束战斗!

事情发展到这里,基本上大家都觉得无趣了,散去。

最后一位小哥Derek Kozel(也是2020 FOSDEM 软件无线电开发者房间主持人,其实出场的人大都现实中熟悉和见过),出来打趣:

I will keep a print out of this tweet in my end-of-PhD party bag. :)

也许这是一次难得的软件无线电、开源硬件、大公司与创客的公开论战。

全文完。

6 阅读 287
推荐阅读
0 条评论
关注数
1350
内容数
21
开源Wi-Fi芯片openwifi项目相关技术进展,欢迎加入
目录
极术微信服务号
关注极术微信号
实时接收点赞提醒和评论通知
Arm中国学堂公众号
关注Arm中国学堂
实时获取免费 Arm 教学资源信息
Arm中国招聘公众号
关注Arm中国招聘
实时获取 Arm 中国职位信息