New Breed of Ultra Low Noise Regulators

Update (01/29/13): Revised the noise density numbers

Update (8/27/12): Added the new TI TPS7A4700 regulator to the table. This device can source up to 1 A of current and ranks on top of the list. Also please ignore the noise density numbers, I need to revise them. Look at the RMS noise values.

The table below compares the noise level of some regulators used in current DIY modules vs a new breed of regulators that are used in portable consumer devices such as cell phones.

Just like phase noise in clocks, it is difficult to compare noise values among linear regulators because there is no common ground in specifying noise figures. Some companies report noise density, others RMS V noise, and yet others % of Vout. The frequency range for the reported noise figures also varies from company to company. Thus it is required to convert spec numbers to a common measuring unit. I chose to convert everything to “Noise Density” in nV/Sqrt(Hz) numbers. In order to understand the relationship between RMS noise and noise density, you can watch this video tutorial: http://www.youtube.com/watch?v=ywChrIRIXWQ [or here]

The noise density numbers are a good approximation and calculated in accordance to the instruction presented in the video above. They have been further normalized to a bandwidth of 100KHz. (I’ve redone the calculation and this time is correct). The number represents the “tail” of the noise spectrum which is typically a flat line.

The first two rows are application notes indicating the noise density value of the noise floor. AN51 is a discrete design using a ZETEX voltage reference. LM723 is an old part used in older designs. Newer monolithic designs have the added advantage that they are also very low dropout (LDO).

The comparison suggests that whether using a discrete design or state of the art monolithic regulators, we are very close to the measurable noise floor (~14 nV seems the best regulators so far…).

For the diyer, these devices are very small and maybe very hard to solder, especially the micro SMD bump package.

Regulator	Used in	Current (mA)	RMS Noise Vout=3.3 (uV)	Bandwidth for spec (Hz)	Noise Density nV/Sqrt(Hz) Normalized to 100KHz
AN124		Noise floor	0.16 (peak)	0-10Hz	1? (not enough data)
AN83		Noise floor	0.5	10-100K	2
TPS7A4700		1000	~4.5	10-100K	14
LP8900		200	6	10-100K	19
AN51		350	3	10-22K	43
LP5900	Lorien XMOS	150	6.5	10-100K	21
LM723		150	2.5	100-10K	79
ADP151	Amanero USB, Ian’s FIFO	200	9	10-100K	28
LT1763, LT1761, LT1762	Buffalo II (LT 1763)	500	20	10-100K	63
TPS79333	Gamma-2	200	32	200-100K	101
LT1963, LT1964	Musiland power mod	1500, 3000	40	10-100K	126
LP2985	Opus DAC	150	30	300-50K	190
TPS786xx		1500	48	100-100K	152
LM340		1000	75	10-100K	237
LM317, LT1117	LCDPS, many low cost devices	1500, 800	99 250	10-10K 10-20K	3131 (?) 3953 (?)

As previously discussed, the noise density numbers are an approximation assuming that it is “flat” throughout the bandwidth of interest. The area under the noise density trace is the RMS noise figure. So basically, if we assume a square area and a bandwidth of 100 KHz, then the RMS Noise = Noise Density X SQRT(100,000) = Noise Density X 316

Note: the second noise figure for the LM317 comes from TNT-Audio [link]. There, it is measured at 250 uV for the ~20Khz bandwidth.

Further reading: http://hifiduino.blogspot.com/2010/03/comparing-noise-figures-in-linear.html

Pray for JAPAN

(Photo from: The Atlantic)

And if you can, GIVE.

Hifiduino in Greece

Here is the project by a reader in Greece using the Arduino Nano (click for larger pictures)

Musiland Working On Mac OS X Driver

From the Musiland Forum:

Q: “那一号，目前乐之邦正在做OSX驱动这件事么？还是说只有意向，啥时候开始做将来看情况？”

“Is Musiland developing an OSX driver or just intends to develop one in the future depending on the situation?”

A: “正在做。”

“Currently developing”

6/1/11: From the Musiland Forum:

Q: “我看乐之邦压根没把linux、mac等小众用户的驱动问题放在心上，以后要寻找其他能在linux下很好驱动的声卡了。或者采用数字输出解码解决”

“I feel Musiland does not have Linux and Mac customers in their mind from the start. People looking for a Linux solution will have to look elsewhere”

A: 不是不放在心上。
你看能在LINUX、MAC用的声卡，动辄都是几千块。
乐之邦的卡买个几百块，那点利润，哪里能支撑那么多的研发人员……
所以，要慢慢来

“It is not that we don’t have those customers in mind. Sound cards for Linux and Mac costs thousands of kuai (implying small market). Musiland sound cards sell for hundreds (implying much larger market). How can we support so much R&D? Therefore slow development”