h********t 发帖数: 555 | 1 it doesn't matter how you sample the data, the quantization noise is
still not band limited, because quantization will introduce high frequency components.the quantization noise spectrum is not relevant to the sampled data spectrum.
In
fs/
you
2.
quantization |
|
c****8 发帖数: 73 | 2 普通数字滤波器一般可以用Matlab来仿真,Matlab有几个函数是可以用来计算
Quantization error的
例如 errvar(q)
It returns the variance of a uniformly distributed random quantization error
that arises from quantizing a signal by quantizer object q.
还有好几个类似的函数,楼主可以做一下research |
|
e**i 发帖数: 249 | 3 这个quantize到底是怎么实现的,比如我有4个bits,我是不是把每个channel先变到0
-1的范围,然后乘以2^bits,然后
floor,再变回0-255的范围?
我按这样quantize后画面暗了很多
order dither我就完全不懂了。。。也没书可查 |
|
h********t 发帖数: 555 | 4 That is right. I do feel it is very diciffult to comnunicate with you.You need understand that the operation of quantization is different from sampling, and the sampling rate of FFT simulation is different from the sampling rate of the data converter under test. If you are interested in how to run FFT to see quantization noise spectrum beyond the data converter sampling rate fs, you can read the book of CMOS Mixed- Signal Circuit Design by Jake Baker. Simulation file and output spetrum are all t |
|
f*********t 发帖数: 37 | 5 Well if 'the sampling rate of FFT test is different from the sampling rate
of the data converter under test', then it makes sense to me.
I am not a circuit design or mixed signal guy. At the input of DSP or
digital ASIC, everything are discrete at sampling rate. Acutally sometimes
they are just some data sitting in memory :). I was talking about the
quantization noise at this point. I think this is also the quantization
noise defined in may DSP books like the one wrote by Proakis. I modeled
thos |
|
G**p 发帖数: 2549 | 6 请问FPGA做DSP运算当信号范围很大时,如何quantization比较好?还有如果scaling?
能不能推荐点资料?谢谢! |
|
f*********t 发帖数: 37 | 7 Sometimes we assume the input of A/D is a white signal and has uniform disitribution between [-max +max]. This is just an assumption. According to this assumption, the quantization noise for each sample will be independent to each other with the same distribtion. Power spectrum of i.i.d random process is white. |
|
f*********t 发帖数: 37 | 8 Not sure if it is related with 滤波器.
Because if the input of A/D is a sine wave, then the quantization noise won'
t be white no matter what filter you use. |
|
le 发帖数: 190 | 9 I understand it's white. The input signal doesn't necessarily need to be
random, it just need to be busy.
I just don't see why its spectrum is from DC to fs/2. Of course, it cannot
be infinity, otherwise the PSD has to be zero.
disitribution between [-max +max]. This is just an assumption. According to
this assumption, the quantization noise for each sample will be independent
to each other with the same distribtion. Power spectrum of i.i.d random
process is white. |
|
h********t 发帖数: 555 | 10 The original question is why quantization noise is band-limited to fs/2. what you claimed is that the power spectrum of the sampled signal from [-fs/2 fs/2] is the same as [(n-1)*fs/2 (n+1)fs/2]. In real world, that will not happen, because that could only happen if you could build an impulse sampling circuit. however, in real world, you just can not implement a impulse sampling circuit. you will have to either use sample and hold or sample and track circuit. in that case, the power spectrum of |
|
h********t 发帖数: 555 | 11 The original question is why quantization noise is band limited to fs/2.what
you claimed is that the spectrum of the sampled data is band limited to fs/
2. From design aspect,as an approximation, we may assume that is true, but
strictly speaking, neither of them is really band limited to fs/2.
with ideal impulse sampling, for the sampled data, the high frequency
spectrum of the original signal will be aliased back to lower frequency,
but the lower frequency components will get high frequency |
|
le 发帖数: 190 | 12 No, quantization error happens after sampling. You cannot use sampling
theory to explain its spectrum distribution.
specified
Discrete
consider
[
other |
|
c*******m 发帖数: 522 | 13 我觉得是看你要保留的bits是多少了。一般quantization noise假设是一个uniform
distribution.数字信号的bits不一样,uniform distribution的variance也不一样。
你可以看下这个书Discrete-time signal processing
[PDF] from ketab.irAV Oppenheim, RW Schafer, JR Buck - 1989
希望有帮助。如果说得不对,还希望各位大牛拍砖 |
|
n********s 发帖数: 150 | 14 I am reading a paper "Class of scalar-field soliton solutions in three space
dimensions" (Phys.Rew.D 13 2739) by T.D.Lee. When the authors try to
quantize the soliton, they said "....Q is the total charge operator. Because
x is a cyclic variable with a period = 2 pi, the eigenvalue of its
conjugate momentum Q are all integers: 0, 1,2....".
I cannot understand this point. What does it means "cyclic variable"? why
its eigenvalue must be integers?
Thanks a lot. |
|
l**p 发帖数: 474 | 15 who know where can I get those standard color quantization images?
such as the fish, parrots, pool....
thanks |
|
w*********g 发帖数: 30882 | 16 高歌,现任北京航空航天大学能源动力学院动力工程及工程热物理学科一级责任教授,
航空发动机气动热力国防重点实验室副主任,长期从事动力工程、工程热物理及流体力
学领域的教学与科研工作,并在基础科研和多学科的应用技术领域取得了一系列国际领
先水平的创新性科研成果。
他在1984年发明的"沙丘驻涡火焰稳定器",获国家发明一等奖,钱学森同志称之为"一
项长中国人志气的重要发明"。该成果广泛应用于我国多种军用航空发动机中,取得了
数以亿元计的经济效益,至今仍保有先进水平。本刊记者于今年10月采访了高歌教授,
了解到了他近期从事的一些前沿科研工作的最新进展,尤其是他对龙卷风的研究及其工
程应用价值,让人耳目一新。
高歌教授在采访中提到,传统的航空发动机技术虽然还在不停地改进提高之中,但
受到原理和材料工艺上的限制,已经逐渐逼近了性能发展的极限。目前虽然涌现出一些
新型航空发动机技术,但仍然没有走出依靠压力膨胀过程来实现热功转换的思路。他强
调,人们应该另辟蹊径,寻找其他可用的工作原理。为此,他研究了自然界龙卷风的形
成与强化机制,发明了一项称为"余热增推"的技术,直接利用龙卷旋涡实现热功转换并
提取... 阅读全帖 |
|
h*****9 发帖数: 6643 | 17 没听说过是你自己无知。这只能说明你是个很容易被骗的无知之徒, 你又要凭你自己的无知骂人,你又无耻了。
揭露 BIG-BANG 是披着科学外衣的宗教,这在10年前就有很多文章了。
THE "BIG BANG" IS JUST RELIGION DISGUISED AS SCIENCE
by Michael Rivero
Once upon a time, a long time ago, there was this guy named Aristotle.
Pretty sharp fellow; he thought up a lot of good things. But, occasionally
he made a mistake.
One mistake he made was to toss an orange up in the air and watch it come
straight back down to his hand. Aristotle reasoned that if he was moving,
the orange would have flown... 阅读全帖 |
|
N*V 发帖数: 350 | 18 【 以下文字转载自 Military 讨论区 】
发信人: wayofflying (小破熊), 信区: Military
标 题: 人类第一台真空能发动机即将在中国攻关成功
发信站: BBS 未名空间站 (Sat May 14 22:28:25 2011, 美东)
高歌,现任北京航空航天大学能源动力学院动力工程及工程热物理学科一级责任教授,
航空发动机气动热力国防重点实验室副主任,长期从事动力工程、工程热物理及流体力
学领域的教学与科研工作,并在基础科研和多学科的应用技术领域取得了一系列国际领
先水平的创新性科研成果。
他在1984年发明的"沙丘驻涡火焰稳定器",获国家发明一等奖,钱学森同志称之为"一
项长中国人志气的重要发明"。该成果广泛应用于我国多种军用航空发动机中,取得了
数以亿元计的经济效益,至今仍保有先进水平。本刊记者于今年10月采访了高歌教授,
了解到了他近期从事的一些前沿科研工作的最新进展,尤其是他对龙卷风的研究及其工
程应用价值,让人耳目一新。
高歌教授在采访中提到,传统的航空发动机技术虽然还在不停地改进提高之中,但
受到原理和材料工艺上的限制,已经逐渐逼近了性能发... 阅读全帖 |
|
m**d 发帖数: 21441 | 19 【 以下文字转载自 Military 讨论区 】
发信人: wayofflying (小破熊), 信区: Military
标 题: 人类第一台真空能发动机即将在中国攻关成功
发信站: BBS 未名空间站 (Sat May 14 22:28:25 2011, 美东)
高歌,现任北京航空航天大学能源动力学院动力工程及工程热物理学科一级责任教授,
航空发动机气动热力国防重点实验室副主任,长期从事动力工程、工程热物理及流体力
学领域的教学与科研工作,并在基础科研和多学科的应用技术领域取得了一系列国际领
先水平的创新性科研成果。
他在1984年发明的"沙丘驻涡火焰稳定器",获国家发明一等奖,钱学森同志称之为"一
项长中国人志气的重要发明"。该成果广泛应用于我国多种军用航空发动机中,取得了
数以亿元计的经济效益,至今仍保有先进水平。本刊记者于今年10月采访了高歌教授,
了解到了他近期从事的一些前沿科研工作的最新进展,尤其是他对龙卷风的研究及其工
程应用价值,让人耳目一新。
高歌教授在采访中提到,传统的航空发动机技术虽然还在不停地改进提高之中,但
受到原理和材料工艺上的限制,已经逐渐逼近了性能发... 阅读全帖 |
|
C********g 发帖数: 9656 | 20 【 以下文字转载自 Military 讨论区 】
发信人: wayofflying (小破熊), 信区: Military
标 题: 人类第一台真空能发动机即将在中国攻关成功
发信站: BBS 未名空间站 (Sat May 14 22:28:25 2011, 美东)
高歌,现任北京航空航天大学能源动力学院动力工程及工程热物理学科一级责任教授,
航空发动机气动热力国防重点实验室副主任,长期从事动力工程、工程热物理及流体力
学领域的教学与科研工作,并在基础科研和多学科的应用技术领域取得了一系列国际领
先水平的创新性科研成果。
他在1984年发明的"沙丘驻涡火焰稳定器",获国家发明一等奖,钱学森同志称之为"一
项长中国人志气的重要发明"。该成果广泛应用于我国多种军用航空发动机中,取得了
数以亿元计的经济效益,至今仍保有先进水平。本刊记者于今年10月采访了高歌教授,
了解到了他近期从事的一些前沿科研工作的最新进展,尤其是他对龙卷风的研究及其工
程应用价值,让人耳目一新。
高歌教授在采访中提到,传统的航空发动机技术虽然还在不停地改进提高之中,但
受到原理和材料工艺上的限制,已经逐渐逼近了性能发... 阅读全帖 |
|
x********g 发帖数: 595 | 21 Because the classical system was first quantized by canonical quantization
in which the Hamiltonian plays the main role.The canonical quantization is
based on Hamilton's equation of motion and the fundamental commutation
relations [p, q] =...( all of which come from the definition of Possion's
bracket).Thus the resulting equations, the Heisenberg's equation of motion
and the Schordinger's equation, contain the Hamiltonian operator.
Another quantization process, the path integral quantization, m |
|
d*******n 发帖数: 369 | 22 For MP1 and MP2, 假设input frame is 512 samples. after it is fed into 32个
bandpass PQMF filter, the output is 512 spectral components grouped into 32
banks or 512 time-domain samples?
Since the input is time-domain samples, the output of the bandpass filters
should be also time-domain samples, right? If so, quantizer will take the
output of the bandpass filters as input but the quantizer needs the input to be spectral components!!! I didn't see an FFT between the filter bank and the quantizer that |
|
c**l 发帖数: 159 | 23 For the digital filter, there are several components contributing to overall
errors.
quantization error of u
quantization error of filter coefficients
computation error in multiplication and adding
Depending on the bandwidth vs sampling ratio of u, there are different
topologies which can be optimized to reduce filter coefficient quantization
error.
Stability of an IIR filter is a different issue, there are bounds that can
guarantee stability. But these bounds are not tight, which typically resu... 阅读全帖 |
|
x********i 发帖数: 905 | 24 http://iccm.mcm.ac.cn/dct/page/1
Invited Lectures
Group 1
Fan Qin: Cluster algebras and monoidal categorification
Fang Li: Positivity of acyclic sign-skew-symmetric cluster algebras via
unfolding method and some related topics
Cheng-Chiang Tsai: An attempt for affine Springer theory
Li Cai: The Gross-Zagier formula: arithmetic applications
Ming-Hsuan Kang: Geometric zeta functions on reductive groups over non-
archimedean local fields
Huanchen Bao: Canonical bases arising... 阅读全帖 |
|
d******0 发帖数: 22800 | 25 ☆─────────────────────────────────────☆
story9988 (小小故事人) 于 (Mon Jan 3 09:24:25 2011, 美东) 提到:
最近经常用到,随便讲一下。
jitter本身就是一个专业术语,可惜,90%的中文书里面看不到正确的解释,10%的英文
专业书,也会有错误的解释。
jitter是 variety of delay。很清楚,很明白,比方说delay一直是10ms,那么就是
constant了,也就是jitter=0.但是,可能么?
这里就解释了为什么有些人提出cd player加个RAM就可以解决的问题,可以么?
不可以!
这也就是为什么jitter消除不了的原因。
再说一下震动,音响器材里面最怕的就是震动,因为器械一震,会影响很多问题,这也
就是脚钉这样一个神奇之物的主要功效。
但是,震动就一定是不好的么?很多时候歪打正着,很可能就是抵消了有害波,这个时
候,就有意思的。
所以了,HIFI为啥这么多年来一直这么有魅力,一句话:不统一的和谐型。
这也是我老见过最有趣的一个东西,一个早就拥有了一个宗旨:高度还原... 阅读全帖 |
|
D**********s 发帖数: 3139 | 26 音量控制对音质的影响,以前有一篇note写得不错,我找了半天才找到,其实还有用变
压器的方法,不过dac这块最保真的方法就是保持位深和标准输出电平,音量控制交给
前级和功放去管就好了,终究绕不过模拟放大这一关的。
Volume Control Technologies
Newsletter Edition:
July 2009 Issue
With the introduction of Benchmark’s HDR-VC™ (High Dynamic Range
Volume Control), many audio enthusiasts are taking another look at volume
control methods. What are the differences between the different volume
control topologies? Is one sonically superior, and why?
The important qualities to consider with respect to volum... 阅读全帖 |
|
s***n 发帖数: 9499 | 27 容易的很,C, N, P三家的JPEG用的Quantization Tables不同,看看文件头就知道了。
Reference:
[1]Digital Image Ballistics from JPEG Quantization
H. Farid, Technical Report, TR2006-583, Dartmouth College, Computer Science
一个著名城市,西雅图。
会打扮,敢打扮,别人都告诉她以后可以考虑当Model。可是阴错阳差,在父母的督促
下,她来到美国的University of Washington就读博士学位。在繁忙的学习中,她尽量
挤出时间来参加社交活动。一
他买这套相机主要是因为看到亚马逊网站上的促销活动,于是抓住机会买个一套。他觉
得500D相对来说轻便,而且有1080P video,ISO可达12800,15M像素也让他觉得很满意
。9点AF和连拍能力虽然差强
因为纯理科的背景,他阅读了大量的材料,最后觉得Nikon应该是更专业的品牌,而且
D90的ISO6400和11点AF应该更厚道。他对像素少,Video能力差,价格偏贵的 |
|
C*********g 发帖数: 3728 | 28 俺一个大陆仔外加非工学院出身的插句嘴:不就是quantization么?现代的
quantization已经发展成颇为复杂的东东,哪是bracketing这么简单的...... |
|
M******r 发帖数: 221 | 29 1)检查一下你的整个loop是stable的吗?如果不stable,snr drop 30dB就很正常了;
2)你从DT转到CT,应该可以算出对应CT的系数;
3)ELP是指从quantizer clock到dac feedback之间的delay,不是整个loop delay。这
个通常做法是
在quantizer之前加个额外的feedback path做compensation;
4)最好不要做high order(>=4) single bit的架构,这个即使你transistor level仿
真的时候可以
稳定,真正tape out回来很容易就不stable了;
30dB |
|
z*****n 发帖数: 7639 | 30 你说的是quantization noise。
他说的有alias存在的情况,不考虑quantization noise,
如果sampling frequency is Fs,对于某一频率f的所有的
高频aliases,包括Fs-f, Fs+f, 2*Fs-f, 2*Fs+f...,都
会被Fs表现为分量f。 |
|
k*********g 发帖数: 791 | 31 quantize space and time 是自变量,怎么quantize? |
|
F******n 发帖数: 160 | 32 2. Origin -- How Did The Problem Arise?
As discussed in previous article (article series #8288, this board), by taking
time as a physical observable, we have to perform the quantization procedure
on "time" variable, such as the canonical quantization method given below:
In classical mechanics, we have canonical variables (H,T) which satisfy the
Hamiltonian equation:
dT/dt={H,T} = 1
where:
H is the system hamiltonian and T its conjugate variable;
"{ }" is the Poisson bracket).
By following the co |
|
g****t 发帖数: 31659 | 33 Approximating a bandlimited function using very coarsely quantized data: A
family of stable sigma-delta modulators of arbitrary order, Annals of Math.,
158 (2)
annals.math.princeton.edu/wp-content/uploads/annals-v158-n2-p09.pdf
请发到z********[email protected],谢谢. |
|
m**********e 发帖数: 12525 | 34 可怜的奥数金牌,知识就永远停留在了大一的数学分析上面,
有时间看看alain connes的大作去吧
发信人: mifepristone (弃婴), 信区: Military
标 题: Re: 时间是连续的吗?
发信站: BBS 未名空间站 (Thu Jan 24 21:19:25 2013, 美东)
一帮外行上窜下跳
这个问题,早已经研究了几十年了,其实就是x,t满足不对易李代数,数学上叫做
noncommutative geometry,如果有兴趣看看数学家的作品:
http://www.alainconnes.org/docs/book94bigpdf.pdf
如果觉得太深奥,也可以看看物理学家的作品:
http://rmp.aps.org/abstract/RMP/v73/i4/p977_1
一句话,因为引力常数太小,所以如果spacetime是quantization的话,李代数的结构
常数不是量子力学里面的planck常数,而是比planck小几十个数量级的常数,所以除非
能达到比big bang的planck scale还高无数倍的能级,否则在理论上讨论都是毫无意义
的... 阅读全帖 |
|
f**x 发帖数: 4325 | 35 define "用", "有用" firstly and quantize them. Then everything will be clear. |
|
|
m********1 发帖数: 681 | 37 还有一点,啥价postulate:
A postulate is an (usually fundamental) assumption a writer makes in order
to discuss a subject in a coherent fashion. Examples of postulates are the
Born rule in quantum mechanics (which defines how the wave function is to be
interpreted), or in classical mechanics the existence of a Lagrangian (
which defines the starting point of theoretical mechanics).
A principle is a more or less universally observed (usually fundamental)
fact. Examples of principles are the second law of... 阅读全帖 |
|
p*******n 发帖数: 1928 | 38 光子是中文的说法,英文里是photon,是光场量子化field quantization后的结果。量
子化后得到一些经典的波动光学无法得到的结论,比如真空中存在真空场。
: 我的意思就是万事皆有它的适用条件和特定定义。
: 这个“静止的光子”是个啥东东啊。
: 我认为既然不以光速运动了,那个粒子的特性已经改变了,不能再叫做“光子”
,它有
: 没有质量和“光子有没有质量”已经不是一个论题了。
|
|
发帖数: 1 | 39 杨振宁先生已经94岁高龄了,近日他公布的《中国今天不宜建造超大对撞机》的意见,
在互联网上引起了热烈的讨论和争议。一位这样高龄的老人还能发声“论战”成为舆论
热点,这种情况也属罕见。于是,对“对撞机”和“高能物理”都不了解的普通人,难
免想要知道,杨振宁为什么会突然在这个问题上“掺和”一下,他是否是这个领域的权
威,他与这次的论战有什么关系?
杨振宁
首先来说明,杨振宁是否权威。
一般公众只知道杨振宁和李政道因“弱相互作用中宇称不受恒”的贡献获得诺贝尔奖。
这项粒子物理领域的成就,已经足以说明杨振宁的权威性了,而实际上这还不是杨振宁
最重要的成果。根据网友的整理,杨振宁有13项世界级的成果,可谓在现代物理的各个
领域全面开花:
(A)统计力学
A1. 1952 Phase Transition(相变理论)。论文序号: 52a,52b, 52c。
A2. 1957 Bosons(玻色子多体问题)。 论文序号: 57h, 57i,57q。
A3. 1967 Yang-Baxter Equation(杨-Baxter方程)。论文序号: 67e。
A4. 1969 Finite Tempera... 阅读全帖 |
|
v**e 发帖数: 8422 | 40 物理学之美:杨振宁的13项重要科学贡献
下面我们分4个领域列出杨振宁的这13项重要贡献以及相关论文在选集中的序号,然后
分别作简要的评述,最后再作进一步的讨论。
http://www.guokr.com/article/442035/
(A)统计力学
A1. 1952 Phase Transition(相变理论)。论文序号: 52a,52b, 52c。
A2. 1957 Bosons(玻色子多体问题): 论文序号: 57h, 57i,57q。
A3. 1967 Yang-Baxter Equation(杨-Baxter方程)。论文序号: 67e。
A4. 1969 Finite Temperature(1维δ函数排斥势中的玻色子在有限温度的严格解)。
论文序号: 69a。
(B) 凝聚态物理
B1. 1961 Flux Quantization(超导体磁通量子化的理论解释)。论文序号: 61c。
B2. 1962 ODLRO(非对角长程序)。论文序号: 62j。
(C) 粒子物理
C1. 1956 Parity Nonconservation (弱相互... 阅读全帖 |
|
d***u 发帖数: 943 | 41 你把均值改成FFT不就能还原了?
要不是FFT会怎么样? 有时能但一定会增加基础噪声,图形的基础噪声quantization是
很重要的一方面 |
|
j*******T 发帖数: 606 | 42 【 以下文字转载自 Dreamer 讨论区 】
发信人: Dreamer (不要问我从哪里来), 信区: Dreamer
标 题: Re: Re: 谁去把FACULTY版的LAZY人肉出来?
发信站: BBS 未名空间站 (Thu Aug 12 18:00:12 2010, 美东)
you should correct your publication list for this one
X. Wang, L. Zhang, ``A High-Quality Medium-Rate Speech Vector Quantizer,''
In Journal of Beijing University of Posts & Telecommunications, Vol. 15, No
1, 1992.
you are not the first author for this paper
http://acad.cnki.net/kns55/detail/detail.aspx?QueryID=3&CurRec=52&DbCode=CJFQ&dbname=CJFD7993&filen |
|
z*****n 发帖数: 7639 | 43 从CD出来的是PCM码流,这个已经是数字信息。
从PCM压制成FLAC或者APE是无损压缩,16bit
到16bit,没有trancation,不存在dithering。
dither用在从模拟到数字转换时候,也就是
生成CD母盘的时候;或者你要把高分辨率比如
16bitPCM信号变成低分辨率信号的时候,你需
要决定如何取舍quantization noise。 |
|
z*****n 发帖数: 7639 | 44 对于蓝光信源(蓝光碟)---蓝光
播放机---hdmi---dac功放这样一套系统来讲,
任何相邻两个设备之间信号传输都存在jitter,
只不过在dac之前都是数字信号,信号也都是
block 编码传输,接收端总可以用一定的buffer
来消除jitter的影响。最后对人来讲,只有
dac的回放时钟jitter对输出产生效果。
不过对于现在的时钟精度(一般都可以做到
44.1k或者其他什么加减几个或者几十个PPM),
这种jitter造成的失真比16比特LPCM的quantization
noise还小。 |
|
b*******e 发帖数: 298 | 45 上上周的onsite interview,是DSP方面的
从早上8:30到下午3:00
一早和2个hr聊一个小时,然后是3个panel的面试,每个panel有2个engineer,中午和2
个manager吃饭,下午又是一个panel,还是蛮累的
记得的一些问题:
FIR v.s. IIR filter
Quantization error influence on IIR
Fixed point arithmetic: Q format,+/-/* of different Q format number
Fixed-point DSP v.s. Floating-point DSP
DSP architecture
Assembly implementation of FIR filtering
Switch a and b without extra variable
Lattice filter
Research projects details
一些体会:
自己简历上的东西要熟
DSP的基本概念要比较清楚
Best wishes for all job-seekers! |
|
b*****o 发帖数: 715 | 46 对于one pass算法,
(1)要求exact median, 可以证明内存需要至少是N/2。
(2)如果允许用随机算法,对于内存量是s,可以先用reservoir sampling采样出s个
sample,然后在对这s个数求median。你可以用概率公式算一下误差。
(3)如果允许approximate median,可以把内存量减少到(log(N))^2。大致想法是把
已经扫描过的数作类似quantization的压缩:对于长度为s的buffer,每个元素不但存一
个value(i),而且还存一个weight(i)。所有的weight加起来就是目前扫描过的数的总数
。算法的关键是要保证:在任何时刻,如果把value(i)重复weight(i)次,然后把所有这
样重复出来的数组成一个stream,用它来计算median,和用原数组计算median,能大致
相当
。细节可参看以下paper:
http://infolab.stanford.edu/~manku/papers/98sigmod-quantiles.pd |
|
f*******w 发帖数: 1243 | 47 Sampling
Quantization
Low-pass/high-pass/notch/comb filters
DFT/FFT
Estimation and detection
Fundamental probability questions |
|
f*******e 发帖数: 4531 | 48 【 以下文字转载自 EE 讨论区 】
发信人: francoise (), 信区: EE
标 题: [Job opening] Audio signal processing engineer @ Qualcomm
发信站: BBS 未名空间站 (Tue Aug 23 15:41:54 2016, 美东)
Please apply online directly.
Job Id E1945678
Job Title Audio Systems Engineer (QCT Audio - San Diego, CA)
Job Overview
Qualcomm multimedia audio systems group develops a broad spectrum of
advanced audio signal processing technologies, including echo cancellation,
noise reduction, array signal processing, acoustic enhancement, voice
en... 阅读全帖 |
|
f*******e 发帖数: 4531 | 49 【 以下文字转载自 EE 讨论区 】
发信人: francoise (), 信区: EE
标 题: [Job opening] Audio signal processing engineer @ Qualcomm
发信站: BBS 未名空间站 (Tue Aug 23 15:41:54 2016, 美东)
Please apply online directly.
Job Id E1945678
Job Title Audio Systems Engineer (QCT Audio - San Diego, CA)
Job Overview
Qualcomm multimedia audio systems group develops a broad spectrum of
advanced audio signal processing technologies, including echo cancellation,
noise reduction, array signal processing, acoustic enhancement, voice
en... 阅读全帖 |
|
f*******e 发帖数: 4531 | 50 Please apply online directly.
Job Id E1945678
Job Title Audio Systems Engineer (QCT Audio - San Diego, CA)
Job Overview
Qualcomm multimedia audio systems group develops a broad spectrum of
advanced audio signal processing technologies, including echo cancellation,
noise reduction, array signal processing, acoustic enhancement, voice
enhancement, active noise cancellation, audio effects, voice activation,
etc
. Such technologies are broadly adopted in mobile phones, IoT (internet of
thing... 阅读全帖 |
|