视频概述
-
-
这里有好多需要说的。让我们从硬件规格开始:
-
Nvidia Tegra X2 (Parker) SoC 配备两颗 Denver 2.0 64-bit 核心和四颗 ARM Cortex A57 64-bit 核心
-
集成了基于 Pascal 的 GPU 配备256个 CUDA 核心
-
8 GB RAM
-
128 GB 板载储存
-
Bluetooth 4.2,Wi-Fi 802.11ac/b/g/n,USB-C,3.5 mm 耳机接口
-
-
-
“混合现实”很复杂,但这和增强你在屏幕上看到的东西其实是一回事(就像智能手机或有外置摄像头反馈的 VR 显示设备)
-
更难的是增加显示,未经过滤的现实直接进入你的眼帘。要摆脱这种不实的错觉,Magic Leap One 使用了几种精巧的技术:
-
波导显示——本质上是一块透明的屏幕从侧面悄无声息的点亮。波导( Magic Leap 称之为“光子光场芯片”)引导光线——在这种情况下,一幅图像,穿过薄薄的一层玻璃,放大并投射入你的眼睛。
-
对焦平面 ——在 VR 显示设备上,一切都是同一焦距的。但在现实则不是这样——有些东西看起来很清晰但其他的看起来是模糊的,取决于你的眼睛的焦点在哪里。Magic Leap 通过合成多个波导来模仿这种现象——将图像分割为清晰的和模糊的区域。
-
-
-
在头带的内侧,我们注意到有一个1级激光产品的标签。在你戴在眼睛前的设备上找到这个标签似乎是件很可怕的事情,但实际上在日常使用中是安全的,并不会比 CD 播放器更加危险。
-
旋开标准的梅花螺丝并移除了盖板,揭示了两个扬声器中的第一个是通过弹簧触点连接的,并通过色彩标记的垫圈保护着——到目前为止,可修复性很强。
-
同样隐藏在这块盖板下面的是:设备唯一的线缆的两个上端和一些帮助调整位置的磁力点。
-
但是从头带右侧突出的那个奇怪的黑色小盒子是什么?
-
-
-
-
拆下其中的一个外部传感器阵列,我们在下面发现了一个光学系统用来将图像传进波导。
-
每个点工作在不同的深度——对应这单层的波导。
-
在背后,我们发现了世纪的显示设备:一个 OmniVision OP02222 场序彩色(FSC)LCOS 设备。这看起来是一个 OmniVision OP02222 的定制版本。
-
-
-
让我们更深入地了解光源和波导光学系统。
-
所以这六层都是干什么的?在两个不同的焦平面上有一个分离的波导来对应每个色彩通道(红绿蓝)。
-
如果没有特定于颜色的波导,每种颜色都会聚焦到稍有不同的点并使图像产生变形。
-
来自 Magic Leap 申请的 2016/0327789 专利的“Figure 6”对我们了解光学器件的内部如何工作有所启发。
-
为了方便你的理解,我们自己为这个系统绘制了一幅“长到没法正常阅读的图像”,但是有猫。
-
-
-
将所有的传感设备连接到头带上,我们得到了一个昂贵的分层柔性电缆:
-
Movidius MA2450 Myriad 2 视觉处理单元
-
SlimPort ANX7530 4K DisplayPort 接收器
-
OmniVision OV680 传感器桥用于同时处理多个相机的图像(就像我们在Amazon Fire 手机上找到的那个一样)
-
Altera/Intel 10M08V81G - 8000 逻辑单元 FPGA,可能用于胶合逻辑,或管理 MV 部件或摄像机桥数据
-
Parade Technologies 8713A 双向 USB 3.0 转接驱动器
-
NXP 半导体 TFA9891 音频放大器
-
德州仪器 TPS65912 电源管理芯片(PMIC)
-
-
-
我们已经享受完光学大餐,是时候把我们的注意力转向这个装置的大脑—— Lightpack!
-
很难不去注意到这个突出的超酷通风口。这太小小的口袋 PC 有没有一个强劲的散热系统?我们马上就能知道。
-
这个 FCC 标记没有放弃自己,其他的就是——由 Magic Leap 设计,在墨西哥组装。事实上的硬件制造厂商的身份据说是一个被严格保密的秘密。
-
-
-
在拆开了几个外壳之后,是时候看看让这魔法发生的芯片们了:
-
两个 Samsung K3RG5G50MM-FGCJ 32 Gb LPDDR4 DRAM(共 64 Gb 或 8 GB )
-
Murata 1KL (像是 Wi-Fi/Bluetooth 模块)
-
Nordic 半导体 N52832 RF SoC
-
Renesas Electronics 9237HRZ 降压 - 升压电池充电器
-
Altera(英特尔拥有)10M08 MAX 10 现场可编程门阵列
-
Maxim Semiconductor MAX77620M 电源管理IC 和 Parade Technologies 8713A双向 USB 3.0转接驱动器
-
由衷感谢以下译者:
74%
这些译者帮助我们修补全世界! 您也想加入吗?
开始翻译 ›
15条评论
In Step 10, figure 2 (the figure from the patent) the ‘optional’ polarizing filter (2072) is not very optional. And you skipped it in your summary image.
It’s likely a three-wavelength, 1/4-wave retarder used to rotate linearly-polarized light to circularly-polarized light before it bounces off the LCOS. The reflected light goes through another 1/4 wave so it’s now 1/2 a wave from the input light. That is - it’s linearly polarized 90 degrees from the original light. That’s why the polarizing beam splitter cube is able to separate the input field (100% coverage) from the LCOS-reflected fields. They’re 90 deg. different polarization.
Also, that means the RGB emitters aren’t likely LED’s. They’d have to be laser diodes (or VCSEL’s) to have the narrow spectrum necessary for the 1/4 wave retarding film.
Wick, you seem to know something about this subject, but you are wrong about the Figure from the patent. The beam splitter will polarize the light but having two might work better.
In the final configuration and as diagrams by iFixit (the 3rd figure with respect to Step 10) they absolutely need a polarizer (as show) right after the LEDs or else half the light will go strait into the injection optics without modulation. This 3rd figure does leave out a quarter-wave plate and polarizing mirror on the right side of the beam splitter cube that is necessary to bounce the light back to the beam splitter so it can be directed out to the injection logic. As drawn, the light from the LCOS would go through the beam splitter and out the side of the cube.
I want to add to my prior comment that I have only seen the teardown pictures and had to mentally reconstruct what was going on from picture. I did not see the components as they were taken off. Based off the pictures, I was trying to make it work based on what I was seen from static pictures.
Something that concerns me in this design is that the LED light is shot straight at the beam splitter and if the polarization is not great with the first/LED polarizer and the beam splitter, there will be a spot of uncontrolled light (about the size of the LEDs) in the center of the image. I’m assuming they are using wire grid polarizers in both locations.
Another issue with the design is that the image has to pass through and reflect off the beam splitter (if based on what I was shown and was told is correct). Usually with a wire grid polarizer, you don’t want to pass the “image” of the LCOS through the polarizer as it will do damage to the image. The FIG 6 from the patents would be the “normal” configuration.
Ye-gods, thats a lot of impressive technology squeezed into a small package! No wonder it has taken so long to develop. I’m skeptical that it will ever be a commercially viable product. They are surely shipping significant dollars along with each developer unit. My prediction is that, at best, it will be a high-end business-use product, never a consumer product. Most probably MagicLeap will be acquired by a major before too long, and a year later the major will kill it, having found the skeletons in the closet and determined it will never be profitable — especially with consumers now embracing lower-tech, lower AR quality devices. But hey, congrats to ML for being aggressive with technology, and I wish them the best of luck. Thanks for the superb tear-down, finally revealing the long-standing mystery of ML, IFIXIT!
lannierose - 回复
The device is reminiscent of a camcorder, with a lot of optics and electronics compressed into a small volume and not a lot of consideration to repairability. There’s a lot more engineering here than a VR headset and I’m impressed how far Magic Leap has iterated the design for a developer release. That said, having a non-replaceable battery would be a disaster for a consumer release IMHO. Great teardown!
Jack Boyce - 回复